A method for determining survival range of salt marsh vegetation based on multi-source data coupling

Abstract

This invention discloses a method for determining the survival range of saline marsh vegetation based on multi-source data coupling. First, a range identification model is constructed to identify the distribution range of saline marsh vegetation. Remote sensing images of the study area are acquired and input into the range identification model to obtain the vegetation distribution in the study area. A hydrodynamic numerical model is established to simulate the tidal flat inundation process, and the tidal flat inundation situation of the entire study area is obtained through the hydrodynamic numerical model. Combining the obtained vegetation distribution and tidal flat inundation situation, a quantitative relationship between tidal flat inundation and vegetation distribution is analyzed. Finally, based on the tidal flat inundation situation of the entire study area, the range within which saline marsh vegetation can survive is determined. The method proposed in this invention overcomes the problem of insufficient spatiotemporal continuity in the tidal flat inundation situation obtained by traditional methods, effectively improving the accuracy of the obtained relationship between tidal flat inundation and vegetation distribution.

Description

Technical Field

[0001] This invention belongs to the field of coastal wetland ecological monitoring technology, specifically involving a method for determining the survival range of salt marsh vegetation based on multi-source data coupling. Background Technology

[0002] Coastal salt marsh wetlands are located at the intersection of land and sea. Salt marsh vegetation includes a variety of plants adapted to the coastal environment, typically exhibiting strong salt and flood tolerance. These plants play an irreplaceable role in blue carbon sequestration, coastal erosion protection, biodiversity maintenance, and coastal wetland ecological restoration. The growth, reproduction, and spatial distribution characteristics of salt marsh vegetation in a given area are all related to the tidal flat inundation situation, including the duration and probability of inundation. Therefore, clarifying the quantitative response relationship between tidal flat inundation and salt marsh vegetation cover in the area, and thus determining the survival range of salt marsh vegetation, is an important scientific basis for coastal wetland ecological protection and restoration, and for the study of tidal flat geomorphological evolution.

[0003] Existing studies on tidal flat inundation and salt marsh vegetation cover primarily rely on remote sensing imagery. For example, Chinese patent ZL202210983756.7, entitled "A Method for Identifying and Estimating the Coverage of Suaeda salsa in the Intertidal Zone of Coastal Areas," proposes a method for identifying the growth density of Suaeda salsa based on the spectral characteristics of remote sensing images. Another example is Chinese patent application No. 202511302879.X, entitled "A Remote Sensing Evaluation Method for Coastal Wetland Ecological Quality under the Background of Spartina alterniflora Expansion," which proposes a method that not only identifies land using remote sensing images but also combines clustering algorithms to identify vegetation distribution on water bodies, thus achieving a more comprehensive ecological quality assessment. While such methods are beneficial for large-scale vegetation cover studies, remote sensing images are limited by various factors, including satellite revisit cycles, cloud cover, and image resolution. They cannot accurately depict the continuous tidal flat inundation situation under detailed topography, thus the survival range of salt marsh vegetation derived from remote sensing images suffers from deficiencies in both computational accuracy and spatiotemporal continuity. In addition, there are two other research methods: using drones to conduct full-coverage aerial surveys of the study area to search for key interfaces of tidal flat land-water interaction, hydrodynamics and vegetation response, and conducting macro-regional statistics on regional inundation probability and vegetation cover to determine the relationship between the two. However, drone aerial surveys are often difficult to design with specific flight paths, resulting in low data acquisition efficiency. Macro-statistical methods cannot determine the detailed topographic inundation situation of tidal flats and the distribution of vegetation cover, leading to insufficient accuracy in determining the survival range of salt marsh vegetation.

[0004] In summary, this paper proposes a method that can efficiently obtain sampling data and, based on the sampling data, determine the tidal flat inundation situation and vegetation cover distribution within the entire geographical area. This method can accurately characterize the relationship between tidal flat inundation and vegetation cover distribution, which is of great significance for determining the accurate survival range of salt marsh vegetation and protecting the coastal wetland ecological environment. Summary of the Invention

[0005] The purpose of this invention is to address the problems of current methods for determining vegetation cover in salt marshes, such as difficulty in collecting data to obtain detailed topographic data on continuous tidal flat inundation processes, or low data collection efficiency, which makes it impossible to efficiently and accurately determine the survival range of vegetation in salt marshes. This invention proposes a method for determining the survival range of vegetation in salt marshes based on multi-source data coupling. This method establishes a hydrodynamic numerical model simulating tidal flat inundation to obtain the continuous tidal flat inundation process, and combines this with vegetation cover distribution to obtain a quantitative response model describing the relationship between inundation probability and vegetation cover.

[0006] The technical solution adopted in this invention is as follows: A method for determining the survival range of vegetation in salt marshes based on multi-source data coupling includes the following steps: S1. Construct a range identification model for recognizing the distribution range of salt marsh vegetation based on image features.

[0007] S2. The area where the survival range of salt marsh vegetation needs to be determined is called the study area. The range of the study area is determined, and a sample period is set. Remote sensing images of the study area are obtained during the sample period. After preprocessing the remote sensing images, they are input into the range recognition model to obtain the vegetation distribution of the study area.

[0008] S3. Establish a hydrodynamic numerical model to simulate the tidal flat inundation process, and use the hydrodynamic numerical model to obtain the tidal flat inundation situation of the entire study area.

[0009] S4. Combining the obtained vegetation distribution and tidal flat inundation, analyze the quantitative relationship between tidal flat inundation and vegetation distribution.

[0010] S5. Based on the quantitative change relationship obtained in step S4 and combined with the tidal flat inundation situation of the entire study area, the range in which the salt marsh vegetation in the study area can survive is determined.

[0011] Existing technologies struggle to efficiently derive complete tidal flat inundation data for an entire study area over a given period. For instance, manual field observations are limited by the difficulty of accessing the tidal flats and the risk of periodic tidal flooding, making long-term, continuous data acquisition difficult. Fixed tide gauges can only capture inundation characteristics at single points, failing to directly extrapolate the probability of continuous inundation across the entire study area. While satellite remote sensing imagery can cover a large area, its revisit period (typically several days) is much longer than the tidal fluctuation cycle, leading to significant sampling bias in the inundation data retrieved from a limited number of remote sensing observations, thus failing to accurately reflect the dynamic characteristics of true tidal flat inundation. In contrast, numerical simulation overcomes the inherent limitations of these methods in terms of spatiotemporal resolution and continuity, providing complete tidal flat inundation data for subsequent research.

[0012] Further optimization involves using a random forest model as the range identification model in step S1. The process of establishing the range identification model specifically includes the following steps: S1.1. Acquire multiple visible light remote sensing images containing salt marsh vegetation growth areas, and the remote sensing images contain various salt marsh vegetation growth areas. Perform preprocessing on all acquired remote sensing images.

[0013] S1.2. The preprocessed remote sensing image is recorded as the initial image. The spectral information of all pixels in the initial image is analyzed. The normalized difference vegetation index (NDVI), normalized difference water index (NDWI), and normalized difference building index (NDBI) of each pixel are calculated one by one based on the spectral information of each pixel. Then, the three index values ​​of the pixel are used as the three-axis coordinates of the RGB color space, so that the corresponding color is obtained through the RGB color space. The obtained color is used as the feature color of the pixel. The feature color of all pixels is obtained according to the above process. Then, the color of all pixels in the initial image is replaced with the feature color to obtain the false color composite image. The false color composite image of all initial images is obtained through the above process.

[0014] S1.3. Establish an initial identification model and set one or more salt marsh vegetation types to be identified. The model takes the features presented by pixels in the false-color composite image and the initial image as input, identifies the pixel type through the presented features, and then obtains the vegetation distribution range by the distribution of pixels identified as vegetation in the image. The pixel type includes vegetation-covered areas and non-vegetated areas, and the features of the vegetation-covered areas are determined based on all vegetation to be identified.

[0015] S1.4. Identify the distribution range of salt marsh vegetation in the false-color composite image, and record the pixels within this range as vegetation pixels; construct a sample set containing multiple training samples, randomly select multiple vegetation pixels as sample pixels in the false-color composite image, each sample pixel corresponds to a training sample, the training sample contains the features presented by the corresponding pixel in the false-color composite image and the initial image respectively, and sample pixels exist in both false-color composite images; randomly select some training samples in the sample set as training data, and the remaining training samples as test data; set discrimination conditions, and train the initial recognition model using the training data, and after each training, obtain the classification results of all test data through the obtained model, and judge whether the model meets the discrimination conditions based on the classification results. If the classification results meet the discrimination conditions, the model is considered to have completed training, and the obtained model is the range recognition model; otherwise, the training is considered incomplete, the training set is optimized, and retraining is performed.

[0016] The formulas for calculating the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Difference Built-up Index (NDBI) for each pixel are as follows: (1) in, NDVI , NDWI , NDBI These are the normalized vegetation index, normalized water index, and normalized building index of the pixel. red The red band reflectance of this pixel. green The green light band reflectance of this pixel. NIR The near-infrared reflectance of this pixel. SWIR This represents the shortwave infrared reflectance of the pixel.

[0017] By completing the training of the range recognition model, the preprocessed remote sensing image is accurately distinguished between salt marsh vegetation and non-target features, thereby identifying the distribution range of salt marsh vegetation.

[0018] Further optimization is needed; the specific discrimination conditions in step S1.4 are as follows: S1.4.1. Number all types that the model can recognize, with each type corresponding to a unique serial number; set two evaluation indicators: Kappa coefficient and overall accuracy. The overall accuracy is the proportion of correctly classified training samples in the image to the total number of training samples. The Kappa coefficient is calculated based on the vegetation distribution using the following formula: (2) in, Kappa Kappa coefficient OA For correct classification coefficients, P e The misclassification coefficient, n The total number of training data, i , j , k All are type serial numbers. M k For real type k Correctly classified as k The number of training samples, M ij For real type i But it is classified as j The number of training samples, M ji For real type j But it was classified as i The number of training samples, l This represents the total number of types.

[0019] S1.4.2. Set a first threshold and a second threshold. If the obtained Kappa coefficient is greater than the first threshold and the overall accuracy is greater than the second threshold, then the model is considered to meet the discrimination condition, training is complete, and the obtained model is the range recognition model; otherwise, training is considered incomplete, the training set is optimized, and retraining is performed. The optimization process is as follows: The confidence level of the training samples is determined based on the distribution location of the corresponding pixels. If a pixel is located at the boundary of the vegetation distribution range, the training sample corresponding to that pixel is of low confidence. After the confidence level of each training sample is determined, all low-confidence training samples in the training set are removed. Then, new pixels that have never been used as sample pixels are selected in the central area of ​​the vegetation range. New training samples are constructed based on the new pixels and added to the training set.

[0020] Further optimization is needed; the process of obtaining vegetation distribution information in step S2 is as follows: The study area was defined, and a sample period was set. Visible light remote sensing images of the study area within the sample period were acquired and preprocessed. The preprocessed remote sensing images were called sampled images. The vegetation distribution range in the sampled images was identified using the range recognition model, and all pixels within the vegetation distribution range were marked. Each marked pixel was also numbered, with each pixel corresponding to a unique serial number. A regression model to describe vegetation cover was established. The regression model was obtained through quadratic polynomial fitting, and the specific formula is as follows: (3) in, m The serial number of the marked cell. Cover m,i For pixels m Inland vegetation type i Coverage (unit: %) a i , b i , c i All are used for calculating vegetation types i The fitting coefficient of the coverage.

[0021] The vegetation cover of each marked pixel is obtained by formula (3), and a vegetation cover raster image is generated based on the vegetation cover data of all pixels.

[0022] Further optimization is achieved in step S3, where the process of determining the tidal flat inundation situation of the entire study area is as follows: S3.1. Based on the vegetation cover distribution in the cover raster image and combined with the land-sea distribution characteristics, determine the raster nodes that can describe the seaward boundary characteristics of the vegetation distribution range. These raster nodes are called boundary nodes. Connect all the boundary nodes in sequence to obtain a continuous boundary line. Conduct aerial surveys of the tidal flat topography of the study area using UAVs. During the aerial survey, the flight path of the UAV is determined according to the boundary line.

[0023] S3.2. Deploy one or more tide level observation stations in the study area. If conditions permit, deploy the tide level observation stations on the boundary line of the study area. Before deployment, divide the boundary line into multiple equal segments, with the number of segments being one more than the number of observation stations. All observation stations correspond one-to-one with all division points, and each tide level observation station is located at the corresponding division point. If it is not possible to deploy them on the boundary line, divide the study area into multiple observation zones. The number of observation zones is equal to the number of observation stations and they correspond one-to-one. Each tide level observation station is located at the geometric center of the corresponding observation zone. During the aerial survey conducted by the UAV, all tide level observation stations remain in monitoring status.

[0024] S3.3. Generate a digital elevation model (DEM) based on the data obtained from aerial surveys to describe the complete terrain of the study area. If the generated DEM model has missing parts, it is necessary to supplement the missing parts by combining elevation data from other sources.

[0025] S3.4. Establish a hydrodynamic numerical model based on the Delft3D simulation system. Based on the hydrodynamic numerical model, combined with the tide monitoring data obtained from the tide observation station and the DEM model, the overall tidal flat inundation situation in the study area is obtained.

[0026] Further optimization is achieved by deriving the tidal flat inundation situation from the hydrodynamic numerical model in step S3.4, as follows: S3.4.1. Based on the geographical features of the study area, set up the computational domain of the Delft3D-FLOW module. The computational domain covers the entire tidal flat area and adjacent sea area of ​​the study area, and establish an orthogonal curve grid covering the entire computational domain. Local densification is performed on the grids located in the vegetation distribution range and near the boundary line.

[0027] S3.4.2. The computational domain is set as an open boundary on the seaward side and the two sides perpendicular to the coastline, driven by the corrected tide level monitoring data; the landward side is set as a closed boundary; the initial conditions of the model are set as still water conditions, and the tide level data at the initial moment is consistent with the tide level monitoring data at the beginning of the monitoring process. S3.4.3. Using the Kriging interpolation method, combined with the elevation data in the DEM model, the elevation of each grid node in the computational domain is obtained, and a topographic field covering the entire computational domain is constructed.

[0028] S3.4.4. Select a region as a sample region in the study area, simulate the tidal flat inundation situation of the sample region over a period of time using a numerical model, and obtain time-varying data of tide level and current velocity based on the simulation results. Then, compare the tide level and current velocity data obtained from the simulation results with the measured data obtained in the same time period in the sample region, and set verification conditions. If the comparison results meet the set verification conditions, the numerical model is considered to meet the verification requirements.

[0029] S3.4.5. Using a model that meets the verification requirements, simulate the tidal flat inundation situation in the study area. The simulation period should be no less than the growth period of Suaeda salsa. Set the distinction criteria between dry and wet states based on the tidal level data. After the calculation is completed, the simulation results include the tidal level data and dry / wet states at all grid nodes at each time step.

[0030] S3.4.6. Number all grid nodes in the model, with each node corresponding to a unique serial number. Calculate the duration of each grid node being submerged by the tide during the simulation period, i.e., the total duration in a wet state. Calculate the submersion probability at each grid node using the following formula: (4) in, q For grid node number, Inundation q For grid nodes q The probability of flooding at a location (in %) N The simulation cycle duration (in days) p For the number of days, T p,q For the first p Tiandian q Duration of flooding (in hours).

[0031] Further optimization is achieved in step S4, where the quantitative relationship between tidal flat inundation and vegetation distribution is derived as follows: S4.1. Obtain the flooding probability raster image based on the flooding probability at each grid node, and then align the flooding probability raster image and the coverage raster image with latitude and longitude coordinates using a projection conversion tool.

[0032] S4.2. Based on the vegetation cover and inundation probability at different locations in the study area, create a dataset to describe the relationship between inundation probability and vegetation cover.

[0033] S4.3. Set a probability step size, divide the 0-100% inundation probability range into multiple intervals according to the set probability step size, and number all intervals sequentially; calculate the mean inundation probability (MFF) and mean vegetation cover (MVF) for each interval using the dataset, and then combine the MFF and MVF of all intervals to derive a quantitative response model describing the MFF-MVF change relationship through fitting. This model is expressed by the following formula: (5) in, r The interval number, MVF i,r For interval r vegetation types i Average coverage MFF r For interval r The average flooding probability, d i , f i , g i All are used for calculating vegetation types i The fitting constant for the average coverage, d i The peak value of the Gaussian curve represents the vegetation type. i The theoretical maximum coverage value, f i The peak value of the Gaussian curve corresponds to the inundation probability, i.e., the vegetation type. i The optimal probability of flooding during growth. g i The bandwidth of the Gaussian curve reflects the vegetation type. i Tolerance range for changes in flooding probability.

[0034] S4.4. Multiple quadrats were set up within the study area, and sample vegetation was selected simultaneously. Each quadrat had a unique number. The measured vegetation cover value of each quadrat was obtained through field measurement, and the location coordinates of each quadrat were determined. Then, the inundation probability of each quadrat was obtained by combining the inundation probability raster image with the location coordinates of the quadrat. The predicted value of sample vegetation cover value of each quadrat was obtained by combining the quantitative response model with the inundation probability. The performance of the quantitative response model was evaluated by the coefficient of determination and the root mean square error. The coefficient of determination and the root mean square error were calculated using the following formulas: (6) in, s The sample plot number is used. M The total number of samples. y s For sample plots s The measured value of vegetation cover in the sample. For sample plots s The predicted value of vegetation cover in the sample. This represents the average measured vegetation cover of all quadrats. R 2 As the coefficient of determination, RMSE This is the root mean square error.

[0035] If a third and fourth threshold are set, and the quantitative response model satisfies... R 2 Not lower than the third threshold at the same time RMSE If the value does not exceed the fourth threshold, the quantitative response model is considered to have been validated. If it does not meet the threshold, the quantitative response model needs to be refitted and validated again through the above process until validation is completed.

[0036] Further optimization is needed. The specific process for determining the survival range of salt marsh vegetation in the study area in step S5 is as follows: The 0-100% inundation probability range was divided into multiple characteristic intervals. The stable growth of vegetation within each interval was determined using a validated quantitative response model. Then, based on the first derivative of the validated quantitative response model with respect to the average inundation probability, the 0-100% inundation probability range was divided into four stages according to the average inundation probabilities corresponding to four scenarios: vegetation cannot survive due to drought, vegetation can survive and its growth is promoted, vegetation can survive but its growth is inhibited, and vegetation cannot survive due to inundation. The three critical points for dividing the inundation probability range were the drought stress critical point, the promotion-inhibition transition point, and the inundation stress critical point. Based on the inundation probability raster map... The inundation probability of each pixel in the image determines its corresponding feature interval, thereby determining the stable growth of vegetation in each pixel. Pixels that can grow stably are recorded as effective pixels. Then, based on three critical points and combined with the inundation probability of each pixel, drought stress critical line, promotion-inhibition transition line, and inundation stress critical line are constructed in the image. The range between the drought stress critical line and the inundation stress critical line in the study area is recorded as the effective growth zone. The effective growth zone is further divided into growth promotion zone and growth inhibition zone by the promotion-inhibition transition line. The area formed by the effective pixels in the effective growth zone is taken as the final survival range of salt marsh vegetation.

[0037] Considering that the habitability of vegetation at different locations on the tidal flat is affected differently by tidal inundation under the influence of topography, and that the drought stress threshold, the promotion-inhibition transition line and the inundation stress threshold need to be constructed in areas where habitability is stable under the influence of tidal forces, otherwise the reliability of the delineated effective growth zone will be low; for pixels within the effective growth zone, if their habitability is not strongly correlated with the influence of tidal forces, the survival probability of vegetation in that area cannot be determined either. Therefore, only the area formed by effective pixels within the effective growth zone can ensure vegetation survival.

[0038] The beneficial effects of the method of the present invention are as follows: 1. The method proposed in this invention determines the distribution of vegetation through remote sensing images, and obtains complete tidal flat inundation information by combining topographic data from UAV aerial surveys and tidal level data measured by tidal level observation stations through a hydrodynamic simulation model. Then, by combining the vegetation distribution and tidal flat inundation information, a quantitative relationship between the two is obtained. This overcomes the problem of insufficient spatiotemporal continuity in the tidal flat inundation information obtained by traditional methods, and effectively improves the accuracy of the obtained relationship between tidal flat inundation and vegetation distribution. Attached Figure Description

[0039] Figure 1 A schematic diagram of the overall process for determining the survival range of vegetation in salt marshes according to the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the method of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below through specific embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0041] Embodiment 1: A method for determining the survival range of salt marsh vegetation based on multi-source data coupling, specifically including the following steps: S1. Construct a range recognition model for identifying the distribution range of salt marsh vegetation according to image features. In this embodiment, the range recognition model is a random forest model. The process of establishing the range recognition model is as follows: S1.1. Obtain multiple visible light remote sensing images of areas where salt marsh vegetation grows. In this embodiment, 5 different locations are selected respectively in six zones, namely the seaward bare flat, seaward sparse zone, landward sparse zone, medium coverage zone, high coverage zone, landward bare flat or vegetation transition zone, to obtain their remote sensing images. The above zones are mainly measured by the coverage (Cover) of Suaeda heteroptera. The definition of each zone is as follows: seaward bare flat, 0% ≤ Cover ≤ 30%, and it is the bare tidal flat at the seaward outer edge of the Suaeda heteroptera community; seaward sparse zone, 30% < Cover ≤ 50%, and it is the transition zone at the seaward edge of the Suaeda heteroptera community; landward sparse zone, 30% < Cover ≤ 50%, and it is the transition zone at the landward edge of the Suaeda heteroptera community; medium coverage zone, 50% < Cover ≤ 80%, and it is the core area inside the Suaeda heteroptera community; high coverage zone, 80% < Cover ≤ 100%, and it is the core area inside the Suaeda heteroptera community; landward bare flat or vegetation transition zone, 0% ≤ Cover ≤ 30%, and it is the bare tidal flat at the landward inner edge of the Suaeda heteroptera community, or the transitional area where it intersects with other vegetation communities. The seaward side is the outer boundary of a zone facing the sea, and the landward side is the inner boundary of a zone facing the land, which is specifically determined by combining the extracted community morphology with the sea-land distribution.

[0042] Then, all remote sensing images are preprocessed. The preprocessing process includes radiometric calibration, atmospheric correction, geometric correction, cloud removal, image cropping, and the unification of the plane coordinate system and elevation datum. The plane coordinate systems of all remote sensing images are uniformly set as the WGS84 coordinate system, and the elevation datum is uniformly set as the 1985 National Elevation Datum.

[0043] S1.2. The preprocessed image is recorded as the initial image. The spectral information of all pixels in the initial image is analyzed. The Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Difference Built-up Index (NDBI) of each pixel are calculated one by one based on the spectral information of each pixel. Then, the three index values ​​of the pixel are used as the three-axis coordinates of the RGB color space, so that the corresponding color is obtained through the RGB color space. The obtained color is the feature color of the pixel. The feature color of all pixels is obtained according to the above process. Then, the color of all pixels in the initial image is replaced with the feature color to obtain the false color composite image. The false color composite image of all initial images is obtained through the above process.

[0044] S1.3. Establish an initial identification model and specify one or more salt marsh vegetation species to be identified. This model uses the features presented by pixels in the false-color composite image and the initial image as input. It identifies pixel types based on these features and then determines the vegetation distribution range by analyzing the distribution of pixels identified as vegetation in the image. The pixel types include vegetation-covered areas and non-vegetated areas. The features of the vegetation-covered areas are jointly determined by all the vegetation species to be identified. In this embodiment, the model uses 30 decision trees, and the remaining hyperparameters use the default settings from the Google Earth Engine platform's Random Forest algorithm application. The salt marsh vegetation to be identified includes three species: reeds, Suaeda salsa, and Spartina alterniflora. The features corresponding to these three species in the false-color composite image and the initial image are listed in Table 1.

[0045] reed Yellow striped / patchy patches Light green Suaeda salsa Dark yellow, fragmented patterns dark green Spartina alterniflora Bright yellow granular / flaky Light green S1.4. Identify the distribution range of salt marsh vegetation in the false-color composite image, and record the pixels within this range as vegetation pixels; construct a sample set containing multiple training samples, randomly select multiple vegetation pixels as sample pixels in the false-color composite image, each sample pixel corresponding to a training sample, the training sample containing the features presented by the corresponding pixel in the false-color composite image and the initial image respectively, and sample pixels exist in both false-color composite images; randomly select a portion of the training samples in the sample set as training data, and the remaining training samples as test data; set discrimination conditions, train the initial recognition model using the training data, and after each training, obtain the classification results of all test data using the obtained model, and judge whether the model meets the discrimination conditions based on the classification results, the specific discrimination conditions are as follows: S1.4.1. During the recognition process, the model classifies all training samples sequentially. There are four categories: reed-covered area, Suaeda salsa-covered area, Spartina alterniflora-covered area, and non-vegetation area. The four types are numbered sequentially. Two evaluation indicators are set: Kappa coefficient and overall accuracy. The overall accuracy is the proportion of correctly classified training samples in the image to the total number of training samples. The Kappa coefficient is calculated according to the vegetation distribution using formula (2). In this embodiment, the total number of types is... l =4.

[0046] S1.4.2. Set a first threshold and a second threshold. If the obtained Kappa coefficient is greater than the first threshold and the overall accuracy is greater than the second threshold, then the model is considered to meet the discrimination condition, training is complete, and the obtained model is the range recognition model; otherwise, training is considered incomplete, the training set is optimized, and retraining is performed. The optimization process is as follows: The confidence level of the training samples is determined based on the distribution location of the corresponding pixels. If a pixel is located at the boundary of the vegetation distribution range, the training sample corresponding to that pixel is of low confidence. After the confidence level of each training sample is determined, all low-confidence training samples in the training set are removed. Then, new pixels that have never been used as sample pixels are selected in the central area of ​​the vegetation range. New training samples are constructed based on the new pixels and added to the training set.

[0047] S2. The area where the survival range of salt marsh vegetation needs to be determined is called the study area. The scope of the study area is determined, and a sample period is set. Visible light remote sensing images of the study area are obtained during the sample period. Then, the remote sensing images are preprocessed as described in step S1.1. In this embodiment, the study area is a typical silty tidal flat coastal wetland. The main species of salt marsh vegetation in this area is Suaeda salsa. The sample period is selected during the period from July to September when Suaeda salsa is typically purplish-red and growing vigorously, and the cloud cover is less than 5%. The remote sensing images are then preprocessed.

[0048] The preprocessed remote sensing image is called the sampled image. The vegetation distribution range in the sampled image is identified by the range recognition model, and all pixels within the vegetation distribution range are marked. At the same time, all marked pixels are numbered, and each pixel corresponds to a unique serial number. A regression model for describing vegetation cover is established. The regression model is obtained by fitting a quadratic polynomial, as shown in formula (3).

[0049] The vegetation cover of each marked pixel is obtained by formula (3), and a vegetation cover raster image is generated based on the vegetation cover data of all pixels.

[0050] S3. Establish a hydrodynamic numerical model to simulate the tidal flat inundation process, and use the hydrodynamic numerical model to obtain the tidal flat inundation situation in the study area, specifically including the following steps: S3.1. Based on the vegetation cover distribution in the vegetation cover raster image and combined with the land-sea distribution characteristics, determine the raster nodes that can describe the seaward boundary characteristics of the vegetation distribution range. These raster nodes are called boundary nodes. Connect all boundary nodes sequentially to obtain a continuous boundary line. Conduct aerial surveys of the tidal flat topography of the study area using a UAV. The flight path of the UAV during the aerial survey is determined according to the boundary line. In this embodiment, the vegetation cover raster images are imported into ArcGIS software. The raster to integer tool in the software is used to convert the vegetation cover distribution raster into integer data to obtain a vegetation distribution map. Then, the surface feature vertex extraction tool is used to obtain the outer boundary nodes of the vegetation community. After obtaining the boundary line, the Bezier smoothing algorithm is used to optimize the inflection points of the boundary line, eliminate sharp angles, ensure the smoothness of the flight path, and generate a KML format flight path file. The drone used was a DJI M350RTK equipped with a Zenmuse L2 LiDAR. The flight path was set at an altitude of 80 m, with 80% forward overlap and 80% lateral overlap. The ground sampling distance was no higher than 3 cm. Aerial surveying was conducted using the LiDAR along the obtained KML format flight path at equal time intervals to acquire LiDAR point cloud data. The survey was conducted during low tide to ensure maximum exposure of the tidal flats and minimize the impact of tide level on the accuracy of the elevation data.

[0051] S3.2. A tide level observation station shall be set up in the study area. If conditions permit, the tide level observation station shall be set up at the midpoint of the boundary line of the study area. Setting up the station at this location can reflect the water level boundary characteristics entering the study area to the greatest extent. If the terrain, such as mudflats, prevents the station from being set up at this location, the tide level observation station shall be set up at the geometric center of the study area. The tide level observation station shall use an RBRsolo3D|wave16 tide gauge to collect time-varying tide level data, with a sampling frequency set to 2 Hz. The tide level observation station shall remain in monitoring status throughout the entire process of UAV aerial survey. After the monitoring is completed, the tide level monitoring data shall be subjected to outlier removal and zero-point drift correction using the 3σ criterion, and the data benchmark shall be unified to the 1985 National Height Datum.

[0052] S3.3. Import the lidar point cloud data obtained from aerial surveys into point cloud processing software. After point cloud filtering and ground point classification, a digital elevation model (DEM) is generated to describe the topography of the study area. In this model, the topographic data is measured using the 1985 National Elevation Datum. If there are missing parts in the generated DEM model, the missing parts are supplemented by publicly available nautical chart water depth data or global topographic data.

[0053] S3.4. A hydrodynamic numerical model was established based on the Delft3D simulation system. The specific process of tidal flat inundation in the study area was obtained through this model as follows: S3.4.1. Based on the geographical features of the study area, set up the computational domain of the Delft3D-FLOW module. The computational domain covers the entire tidal flat area and adjacent sea area of ​​the study area, and establish an orthogonal curve grid covering the entire computational domain. Local densification is performed on the grids located in the vegetation distribution range and near the boundary line.

[0054] S3.4.2. The computational domain is set as an open boundary on the seaward side and the two sides perpendicular to the coastline, driven by the corrected tide level monitoring data; the landward side is set as a closed boundary; the initial conditions of the model are set as still water conditions, and the tide level data at the initial moment is consistent with the tide level monitoring data at the beginning of the monitoring process. S3.4.3. Using the Kriging interpolation method, combined with the elevation data in the DEM model, the elevation of each grid node in the computational domain is obtained, and a topographic field covering the entire computational domain is constructed.

[0055] S3.4.4. Select a region as a sample area within the study area, and simulate the tidal flat inundation situation in the sample area over a period of time using a numerical model. Obtain time-varying data of tidal level and current velocity based on the simulation results. Then, compare the tidal level and current velocity data obtained from the simulation results with the measured data obtained in the same time period in the sample area. Calculate the root mean square error of the tidal level data at all times, and compare the time-varying trend of the current velocity data. If the root mean square error does not exceed 0.15 m, and the trend of the simulated and measured current velocity data is basically consistent, then the numerical model is considered to meet the verification requirements.

[0056] S3.4.5 The inundation of the tidal flats in the study area is simulated using a model that meets the verification requirements. The simulation period is no less than the growth period of Suaeda salsa, which is set to 90 days in this embodiment. The time step is set to 15 s. After the calculation is completed, the simulation results are obtained by taking the tidal level data and dry / wet status at all grid nodes at each time step. In this embodiment, if the tidal level at a certain location does not exceed 0.005 m, it is considered to be in a dry state; if it exceeds 0.01 m, it is considered to be in a wet state. In this way, the simulation results can completely capture the alternation of dry and wet tidal flats and ensure the spatiotemporal continuity of the inundation probability calculation.

[0057] S3.4.6 Number all grid nodes in the model, with each node corresponding to a unique serial number. Calculate the duration of each grid node being submerged by the tide during the simulation period, i.e. the total duration in a wet state, and calculate the submersion probability at each grid node using the following formula (4).

[0058] S4. By comparing and analyzing the vegetation cover raster images of the study area with the tidal flat inundation situation, the relationship between inundation probability and vegetation cover was obtained. The specific process is as follows: S4.1. Obtain the flooding probability raster image based on the flooding probability at each grid node, and then use a projection conversion tool to align the flooding probability raster image and the coverage raster image with latitude and longitude coordinates to eliminate spatial offset errors.

[0059] S4.2. Based on the vegetation cover and inundation probability at each grid node of the inundation probability raster image, a dataset describing the correspondence between inundation probability and vegetation cover is created. In this embodiment, the spatial analysis tools of ArcGIS software are used to create the dataset.

[0060] S4.3. Divide the 0-100% inundation probability range into 100 intervals with a step size of 1%, and number all intervals sequentially; calculate the mean inundation probability (MFF) and mean vegetation cover (MVF) of each interval using the dataset, and then use the Gaussian nonlinear fitting method to combine the MFF and MVF of all intervals to obtain a quantitative response model to describe the relationship between MFF and MVF. In the fitting process, the least squares method is used to iterate and obtain the optimal solution of the fitting constant. The model is expressed by the following formula (5).

[0061] S4.4. Multiple quadrats were set up in the study area, each with a unique number. The measured value of Suaeda salsa coverage of each quadrat was obtained through field measurement, and the location coordinates of each quadrat were determined. Then, the flooding probability of each quadrat was obtained by combining the flooding probability raster image with the location coordinates of the quadrat. The predicted value of Suaeda salsa coverage of each quadrat was obtained by combining the flooding probability with the quantitative response model. The performance of the quantitative response model was evaluated by the coefficient of determination and the root mean square error. The coefficient of determination and the root mean square error were calculated by the following formula (6).

[0062] If a third and fourth threshold are set, and the quantitative response model satisfies... R 2 Not lower than the third threshold at the same time RMSE If the value does not exceed the fourth threshold, the quantitative response model is considered to have been validated. If it does not meet the threshold, the quantitative response model needs to be refitted and validated again through the above process until validation is completed.

[0063] S5. The 0-100% inundation probability range is divided into five characteristic intervals: 0-10%, 10%-30%, 30%-60%, 60%-80%, and >80%. The mean, maximum, minimum, and standard deviation of vegetation cover within each characteristic interval are obtained using a validated quantitative response model to quantify the stable growth of vegetation in different characteristic intervals. The vegetation cover is primarily based on the cover of *Suaeda salsa*, combined with the cover of two other vegetation types. Then, based on the first derivative of the validated quantitative response model with respect to the average inundation probability, the 0-100% inundation probability range is divided into four stages according to the average inundation probabilities corresponding to four scenarios: vegetation cannot survive due to drought, vegetation can survive and its growth is promoted, vegetation can survive but its growth is inhibited, and vegetation cannot survive due to inundation. The three critical points for dividing the inundation probability range are the drought stress critical point, the promotion-inhibition transition point, and the inundation stress critical point. The critical point of drought stress is determined by identifying the corresponding feature interval of each pixel in the inundation probability raster image. This determines the stable growth of vegetation in each pixel, and pixels that can grow stably are recorded as effective pixels. Then, based on the drought stress critical point, the inundation stress critical point, and the promotion-inhibition transition point, combined with the inundation probability of each pixel, drought stress critical lines, promotion-inhibition transition lines, and inundation stress critical lines are constructed in the image. The area between the drought stress critical line and the inundation stress critical line in the study area is recorded as the effective growth zone. The effective growth zone is further divided into growth promotion zone and growth inhibition zone by the promotion-inhibition transition line. The area formed by the effective pixels in the effective growth zone is taken as the final survival range of the salt marsh vegetation.

Claims

1. A method for determining the survival range of vegetation in salt marshes based on multi-source data coupling, characterized in that, Specifically, the following steps are included: S1. Construct a range identification model for recognizing the distribution range of salt marsh vegetation based on image features; S2. The area where the survival range of salt marsh vegetation needs to be determined is called the study area. The range of the study area is determined, and a sample period is set. Remote sensing images of the study area are obtained during the sample period. The remote sensing images are preprocessed and then input into the range recognition model to obtain the vegetation distribution of the study area. S3. Establish a hydrodynamic numerical model to simulate the tidal flat inundation process, and use the hydrodynamic numerical model to obtain the tidal flat inundation situation of the entire study area; S4. Combining the obtained vegetation distribution and tidal flat inundation, analyze the quantitative relationship between tidal flat inundation and vegetation distribution; S5. Based on the quantitative change relationship obtained in step S4, and combined with the tidal flat inundation situation of the entire study area, the range within which the salt marsh vegetation in the study area can survive is determined.

2. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 1, characterized in that, The range identification model mentioned in step S1 is a random forest model. The process of establishing the range identification model specifically includes the following steps: S1.

1. Acquire multiple visible light remote sensing images containing salt marsh vegetation growth areas, and the remote sensing images contain various salt marsh vegetation growth areas. Perform preprocessing on all acquired remote sensing images. S1.

2. The preprocessed remote sensing image is recorded as the initial image. The spectral information of all pixels in the initial image is analyzed. The normalized difference vegetation index (NDVI), normalized difference water index (NDWI), and normalized difference building index (NDBI) of each pixel are calculated one by one based on the spectral information of each pixel. Then, the three index values ​​of the pixel are used as the three-axis coordinates of the RGB color space, so that the corresponding color is obtained through the RGB color space. The obtained color is used as the feature color of the pixel. The feature color of all pixels is obtained according to the above process. Then, the color of all pixels in the initial image is replaced with the feature color to obtain the false color composite image. The false color composite image of all initial images is obtained through the above process. S1.

3. Establish an initial identification model and set one or more salt marsh vegetation species to be identified. The model takes the features presented by pixels in the false-color composite image and the initial image as input, identifies the pixel type through the presented features, and then obtains the vegetation distribution range by the distribution of pixels identified as vegetation in the image. The pixel type includes vegetation-covered areas and non-vegetated areas, and the features of the vegetation-covered areas are determined based on all the vegetation to be identified. S1.

4. Identify the distribution range of salt marsh vegetation in the false-color composite image, and record the pixels within this range as vegetation pixels; construct a sample set containing multiple training samples, randomly select multiple vegetation pixels as sample pixels in the false-color composite image, each sample pixel corresponds to a training sample, the training sample contains the features presented by the corresponding pixel in the false-color composite image and the initial image respectively, and sample pixels exist in both false-color composite images; randomly select some training samples in the sample set as training data, and the remaining training samples as test data; set discrimination conditions, and train the initial recognition model using the training data, and after each training, obtain the classification results of all test data through the obtained model, and judge whether the model meets the discrimination conditions based on the classification results. If the classification results meet the discrimination conditions, the model is considered to have completed training, and the obtained model is the range recognition model; otherwise, the training is considered incomplete, the training set is optimized, and retraining is performed.

3. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 2, characterized in that, The specific discrimination conditions described in step S1.4 are as follows: S1.4.

1. Number all types that the model can recognize, with each type corresponding to a unique serial number; set two evaluation indicators: Kappa coefficient and overall accuracy. The overall accuracy is the proportion of correctly classified training samples in the image to the total number of training samples. The Kappa coefficient is calculated based on the vegetation distribution using the following formula: (1) in, Kappa Kappa coefficient OA For correct classification coefficients, P e The misclassification coefficient, n For the total number of training data, i , j , k All are type serial numbers. M k For real type k Correctly classified as k The number of training samples, M ij For real type i But it is classified as j The number of training samples, M ji For real type j But it is classified as i The number of training samples, l Total number of types; S1.4.

2. Set a first threshold and a second threshold. If the obtained Kappa coefficient is greater than the first threshold and the overall accuracy is greater than the second threshold, then the model is considered to meet the discrimination condition, training is complete, and the obtained model is the range recognition model; otherwise, training is considered incomplete, the training set is optimized, and retraining is performed. The optimization process is as follows: The confidence level of the training samples is determined based on the distribution location of the corresponding pixels. If a pixel is located at the boundary of the vegetation distribution range, the training sample corresponding to that pixel is of low confidence. After the confidence level of each training sample is determined, all low-confidence training samples in the training set are removed. Then, new pixels that have never been used as sample pixels are selected in the central area of ​​the vegetation range. New training samples are constructed based on the new pixels and added to the training set.

4. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 1, characterized in that, The process of obtaining the vegetation distribution in step S2 is as follows: The study area was defined, and a sample period was set. Visible light remote sensing images of the study area within the sample period were acquired and preprocessed. The preprocessed remote sensing images were called sampled images. The vegetation distribution range in the sampled images was identified using the range recognition model, and all pixels within the vegetation distribution range were marked. Each marked pixel was also numbered, with each pixel corresponding to a unique serial number. A regression model to describe vegetation cover was established. The regression model was obtained through quadratic polynomial fitting, and the specific formula is as follows: (2) in, m The serial number of the marked cell. Cover m,i For pixels m Inland vegetation type i Coverage (unit: %) a i , b i , c i All are used for calculating vegetation types i The fitting coefficient of the coverage; The vegetation cover of each marked pixel is obtained by formula (2), and a vegetation cover raster image is generated based on the vegetation cover data of all pixels.

5. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 1, characterized in that, The process of obtaining the tidal flat inundation status of the entire study area in step S3 is as follows: S3.

1. Based on the vegetation cover distribution in the vegetation cover raster image and combined with the land-sea distribution characteristics, determine the raster nodes that can describe the seaward boundary characteristics of the vegetation distribution range. These raster nodes are called boundary nodes. Connect all the boundary nodes in sequence to obtain a continuous boundary line. Conduct aerial surveys of the tidal flat topography of the study area using UAVs. During the aerial survey, the flight path of the UAV is determined according to the boundary line. S3.

2. Deploy one or more tide level observation stations in the study area. If conditions permit, deploy the tide level observation stations on the boundary line of the study area. Before deployment, divide the boundary line into multiple equal segments, with the number of segments being one more than the number of observation stations. All observation stations correspond one-to-one with all division points, and each tide level observation station is located at its corresponding division point. If it is not possible to deploy them on the boundary line, divide the study area into multiple observation zones. The number of observation zones is equal to the number of observation stations and they correspond one-to-one. Each tide level observation station is located at the geometric center of its corresponding observation zone. During the aerial survey conducted by the UAV, all tide level observation stations remain in monitoring status. S3.

3. Generate a digital elevation model (DEM) to describe the complete terrain of the study area based on the data obtained from aerial surveys. If the generated DEM model has missing parts, it is necessary to supplement the missing parts by combining elevation data from other sources. S3.

4. Establish a hydrodynamic numerical model based on the Delft3D simulation system. Based on the hydrodynamic numerical model, combined with the tide monitoring data obtained from the tide observation station and the DEM model, the overall tidal flat inundation situation in the study area is obtained.

6. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 5, characterized in that, The process of obtaining the tidal flat inundation situation through the hydrodynamic numerical model in step S3.4 is as follows: S3.4.

1. Based on the geographical features of the study area, the Delft3D-FLOW module computation domain is set up. The computation domain covers the entire tidal flat area and adjacent sea area of ​​the study area, and an orthogonal curve grid covering the entire computation domain is established. The grids located in the vegetation distribution range and near the boundary line are locally refined. S3.4.

2. The computational domain is set as an open boundary on the seaward side and the two sides perpendicular to the coastline, driven by the corrected tide level monitoring data; the landward side is set as a closed boundary; the initial conditions of the model are set as still water conditions, and the tide level data at the initial moment is consistent with the tide level monitoring data at the beginning of the monitoring process. S3.4.

3. Using the Kriging interpolation method and combining the elevation data in the DEM model, the elevation of each grid node in the computational domain is obtained, and a topographic field covering the entire computational domain is constructed. S3.4.

4. Select a region as a sample region in the study area, simulate the tidal flat inundation situation of the sample region over a period of time using a numerical model, and obtain time-varying data of tide level and flow velocity based on the simulation results. Then, compare the tide level and flow velocity data obtained from the simulation results with the measured data obtained in the same time period in the sample region, and set verification conditions. If the comparison results meet the set verification conditions, the numerical model is considered to meet the verification requirements. S3.4.

5. Use the model that meets the verification requirements to simulate the inundation of tidal flats in the study area. The simulation period shall not be less than the growth period of Suaeda salsa. Set the distinction between dry and wet states according to the tidal data. The simulation results obtained after the calculation include the tidal data and dry / wet states at all grid nodes at each time step. S3.4.

6. Number all grid nodes in the model, with each node corresponding to a unique serial number. Calculate the duration of each grid node being submerged by the tide during the simulation period, i.e., the total duration in a wet state. Calculate the submersion probability at each grid node using the following formula: (3) in, q For grid node number, Inundation q For grid nodes q The probability of flooding at a location (in %) N The simulation cycle duration (in days) p For the number of days, T p,q For the first p Tiandian q Duration of flooding (in hours).

7. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 1, characterized in that, The process of obtaining the quantitative relationship between tidal flat inundation and vegetation distribution in step S4 is as follows: S4.

1. Obtain the flooding probability raster image based on the flooding probability at each grid node, and then align the flooding probability raster image and the coverage raster image with latitude and longitude coordinates using a projection transformation tool; S4.

2. Based on the vegetation cover and inundation probability at different locations in the study area, create a dataset to describe the relationship between inundation probability and vegetation cover; S4.

3. Set a probability step size, divide the 0-100% inundation probability range into multiple intervals according to the set probability step size, and number all intervals sequentially; calculate the mean inundation probability (MFF) and mean vegetation cover (MVF) for each interval using the dataset, and then combine the MFF and MVF of all intervals to derive a quantitative response model describing the MFF-MVF change relationship through fitting. This model is expressed by the following formula: (4) in, r The interval number, MVF i,r For interval r vegetation types i Average coverage MFF r For interval r The average flooding probability, d i , f i , g i All are used for calculating vegetation types i The fitting constant for the average coverage, d i The peak value of the Gaussian curve represents the vegetation type. i The theoretical maximum coverage value, f i The peak value of the Gaussian curve corresponds to the inundation probability, i.e., the vegetation type. i The optimal probability of flooding during growth. g i The bandwidth of the Gaussian curve reflects the vegetation type. i Tolerance range to changes in flooding probability; S4.

4. Multiple quadrats were set up within the study area, and sample vegetation was selected simultaneously. Each quadrat had a unique number. The measured vegetation cover value of each quadrat was obtained through field measurement, and the location coordinates of each quadrat were determined. Then, the inundation probability of each quadrat was obtained by combining the inundation probability raster image with the location coordinates of the quadrat. The predicted value of sample vegetation cover value of each quadrat was obtained by combining the quantitative response model with the inundation probability. The performance of the quantitative response model was evaluated by the coefficient of determination and the root mean square error. The coefficient of determination and the root mean square error were calculated using the following formulas: (5) in, s The sample plot number is used. M The total number of samples. y s For sample plots s The measured value of vegetation cover in the sample. For sample plots s The predicted value of vegetation cover in the sample. This represents the average measured vegetation cover of all quadrats. R 2 As the coefficient of determination, RMSE This is the root mean square error; If a third and fourth threshold are set, and the quantitative response model satisfies... R 2 Not lower than the third threshold at the same time RMSE If the value does not exceed the fourth threshold, the quantitative response model is considered to have been validated. If it does not meet the threshold, the quantitative response model needs to be refitted and validated again through the above process until validation is completed.

8. The method for determining the survival range of saline marsh vegetation based on multi-source data coupling as described in claim 1, characterized in that, The process of determining the survival range of salt marsh vegetation in the study area in step S5 is as follows: The 0-100% inundation probability range was divided into multiple characteristic intervals. The stable growth of vegetation within each interval was determined using a validated quantitative response model. Then, based on the first derivative of the validated quantitative response model with respect to the average inundation probability, the 0-100% inundation probability range was divided into four stages according to the average inundation probabilities corresponding to four scenarios: vegetation cannot survive due to drought, vegetation can survive and its growth is promoted, vegetation can survive but its growth is inhibited, and vegetation cannot survive due to inundation. The three critical points for dividing the inundation probability range were the drought stress critical point, the promotion-inhibition transition point, and the inundation stress critical point. Based on the inundation probability raster map... The inundation probability of each pixel in the image determines its corresponding feature interval, thereby determining the stable growth of vegetation in each pixel. Pixels that can grow stably are recorded as effective pixels. Then, based on three critical points and combined with the inundation probability of each pixel, drought stress critical line, promotion-inhibition transition line, and inundation stress critical line are constructed in the image. The range between the drought stress critical line and the inundation stress critical line in the study area is recorded as the effective growth zone. The effective growth zone is further divided into growth promotion zone and growth inhibition zone by the promotion-inhibition transition line. The area formed by the effective pixels in the effective growth zone is taken as the final survival range of salt marsh vegetation.

Images