Configuration method suitable for water demand of plateau wetland landscape ecology

Abstract

The invention discloses a configuration method suitable for plateau wetland landscape ecological water demand, and the method is characterized in that the method comprises the following steps: S1, obtaining a remote sensing image, carrying out the image classification through a random forest algorithm based on a particle swarm optimization algorithm, and generating a landscape type classification graph; s2, calculating a landscape ecological risk index, identifying a sudden change point of the flow to the landscape ecological risk index by using non-parametric change point analysis, and determining an initial flow and a peak flow of each vegetation growth period; s3, selecting hydrological elements with ecological significance, and quantitatively analyzing wetland hydrological situation changes by adopting ecological hydrological indexes; and S4, coupling an ecological hydrological process, determining a landscape ecological water demand scheme of each time node in the vegetation growth season, and drawing a landscape ecological water demand process. According to the method, the influence of ecological hydrological index changes on the landscape pattern and the ecological risk can be analyzed, and the landscape ecological water demand of the plateau river wetland is quantitatively calculated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water environment ecology, and in particular to a configuration method suitable for plateau wetland landscape ecological water demand. BACKGROUND

[0002] The plateau wetland is jointly constructed by a complex and fragile ecological network of its cold climate, variable dry and wet conditions, short vegetation growth period, thin atmospheric oxygen content, and frequent natural disasters. In recent years, with the rapid development of economic society and the continuous growth of population, the land use type of the plateau wetland has undergone great changes, leading to changes in the landscape pattern, thus causing changes in the ecological hydrological process and the ecological water demand pattern. In addition, the landscape water demand is one of the key factors affecting the economic development of the plateau wetland, and it is of great significance to the water cycle utilization in the region. However, at present, the landscape pattern characteristics of the ecological sensitive area with the landscape as the protection target are not clear, the coupling relationship between the landscape ecology and the ecological hydrological index of the water flow characteristics is lacking, it is difficult to accurately assess the influence of water diversion on the river wetland landscape and the river continuity, the quantitative analysis basis is lacking, and a multi-target, multi-process river wetland ecological water demand evaluation system has not been formed.

[0003] There are many progresses in the coupling research of landscape pattern and hydrological regime in the existing landscape ecological water demand acquisition method, but the response mechanism of landscape pattern evolution, the correlation analysis between landscape pattern characteristics and ecological-hydrological elements, and the special plant growth rhythm of the plateau wetland are not clear. With the development of new technologies such as remote sensing and GIS, the mutual influence and mutual influence between wetland landscape and hydrology-topography have become the research focus at home and abroad, so it is necessary to develop a more suitable configuration system for the plateau wetland landscape ecological water demand based on remote sensing data interpretation, so as to maintain the health and stability of the wetland ecosystem. SUMMARY

[0004] In order to solve the problems of the existing landscape ecological water demand acquisition method, such as lack of response mechanism of landscape pattern evolution, correlation analysis between landscape pattern characteristics and ecological-hydrological elements, and consideration of plateau wetland landscape ecological water demand, the present application proposes a configuration method suitable for plateau wetland landscape ecological water demand, which solves the above problems.

[0005] The present application discloses a configuration method suitable for plateau wetland landscape ecological water demand, which comprises the following steps:

[0006] S1, acquiring a remote sensing image, using a random forest algorithm based on a particle swarm optimization algorithm to classify the image, generating a landscape type classification map, and performing precision test and post-processing on the classification result;

[0007] S2, calculate the landscape ecological risk index by coupling the plateau wetland landscape pattern index, identify the mutation point of the landscape ecological risk index by using non-parametric variable point analysis, and determine the initial flow and peak flow of each vegetation growing season;

[0008] S3, select hydrological elements with ecological significance, and use ecological hydrological index to quantitatively analyze the change of wetland hydrological regime;

[0009] S4, coupling ecological hydrological process, determining the landscape ecological water demand scheme of each time node in the vegetation growing season, and drawing the landscape ecological water demand process.

[0010] Preferably, according to the configuration method suitable for plateau wetland landscape ecological water demand of claim 1, wherein the S1 comprises the following steps:

[0011] S11, download Sentinel-2 L2A level cloud-free remote sensing image data, and extract sample points in the study area as a training set;

[0012] S12, extract spectral features, texture features, geometric features and context features from the remote sensing image to construct an initial feature set;

[0013] Spectral features are the most basic and commonly used features in remote sensing image classification, which reflect the reflection and radiation characteristics of ground objects in different bands. By using the reflectivity difference between red band and near-infrared band, normalized vegetation index and enhanced vegetation index are used to distinguish vegetation and non-vegetation areas. By using the reflectivity difference between green band and near-infrared band, normalized water index and enhanced water index are used to distinguish water and non-water areas;

[0014] Texture features describe the spatial variation pattern of gray scale distribution in the image, which is an important parameter reflecting the roughness and structural complexity of the surface of the ground object. By local binary pattern analysis method, the gray scale values of the center pixel and its neighborhood pixels are compared, binary pattern code is generated, statistical quantities are constructed based on gray level co-occurrence matrix as texture classification features, and wavelet transform is used to multi-scale decomposition of the image, texture information in different scales is extracted, and details and edge features in the image are captured, so as to effectively assist in distinguishing different landscape types with similar spectral features;

[0015] Geometric features describe the shape, size and structure of the ground object, which is an important basis for distinguishing different ground object types. By calculating the area and perimeter of the image object, different types of ground objects are preliminarily distinguished, and by calculating the circularity to reflect the degree of object close to circle, calculating the aspect ratio to describe the length-width ratio of object, and calculating the compactness to measure the smoothness of object boundary;

[0016] Contextual features describe the spatial relationship and adjacency characteristics between image objects, which is an important means to further improve the classification accuracy. By calculating the adjacency matrix between image objects, the adjacency features and spatial layout information of the objects are obtained, so as to more accurately identify the types of ground objects in complex scenes;

[0017] S13, using a particle swarm optimization algorithm to optimize the hyperparameters of the random forest;

[0018] S14, using the optimized random forest algorithm to evaluate the initial feature set, calculating the importance of the features, and constructing a feature subset according to the importance of the features;

[0019] S15, at each classification node of the hierarchical classification tree, using the selected feature subset to train the random forest classifier and constructing a classification model;

[0020] S16, inputting the image objects to be classified into the hierarchical classification tree, and classifying and predicting layer by layer according to the classification rules and feature set to obtain the landscape type classification result;

[0021] S17, in the interpretation process, for the classes that are easy to confuse, after being identified by visual interpretation, the high-definition resolution satellite image of GoogleEarth Pro is used for verification, and finally the plateau wetland landscape is divided into 5 categories of beach, farmland, forest land, grassland and water area, and a detailed landscape pattern distribution map is generated;

[0022] S18, using the confusion matrix to evaluate the accuracy of the classification result, and using the indicators including user accuracy, producer accuracy, overall accuracy and Kappa coefficient to comprehensively evaluate the accuracy of the classification result.

[0023] The user accuracy calculation formula is:

[0024]

[0025] The producer accuracy calculation formula is:

[0026]

[0027] The overall accuracy calculation formula is:

[0028]

[0029] Wherein, M ii is the diagonal element of the confusion matrix (i.e. the number of correctly classified samples), is the number of all samples predicted by the model as class i, and j is the other possible class, is the number of samples actually belonging to class i, and j is the other possible class, is the number of all samples.

[0030] The formula for calculating the kappa coefficient is:

[0031]

[0032] where p0 is the overall accuracy, p e is the expected accuracy of random classification, k is the total number of classes, and N is the total number of samples.

[0033] Preferably, the S13 comprises the following steps:

[0034] S131, randomly initializing a group of particles in the hyperparameter space, each particle representing a possible combination of parameters;

[0035] S132, updating the speed and position of each particle according to its current speed, position, individual historical optimal position, and global optimal position;

[0036] S133, using the performance indicator of the random forest classifier on the training set as the fitness function to calculate the fitness value of each particle;

[0037] S134, updating the individual optimal position of each particle and the global optimal position of the entire particle group according to the fitness value;

[0038] S135, repeating S131-S134 until the preset stopping condition is met.

[0039] Preferably, the S14 comprises the following steps:

[0040] S141, using the random forest algorithm to calculate the impact of each feature on performance during training, thereby obtaining the importance of the feature;

[0041] S142, sorting the features according to their importance;

[0042] S143, selecting the top N most important features to construct an optimal feature subset.

[0043] Preferably, the S15 comprises the following steps:

[0044] S151, using the optimal feature subset and the corresponding data subset to train the random forest classifier at each node;

[0045] S152, evaluating the random forest classifier by calculating the evaluation indicator to check whether its performance meets the requirements;

[0046] S153, connecting the trained random forest classifier to the corresponding node of the hierarchical classification tree to construct a classification model;

[0047] S154, optimize the classification model by adjusting the hierarchy, reselecting the feature subset, and optimizing the random forest parameters, and improve the overall performance of the classification model.

[0048] Preferably, S2 comprises the following steps:

[0049] S21, further coupled with the calculation of the landscape ecological risk index by calculating the landscape pattern index, identify the high-risk area in the river wetland, the calculation formula of the landscape ecological risk index is:

[0050]

[0051] Wherein, A ki is the area of land class i in the kth risk plot, A k is the area of the kth risk plot, R i is the landscape loss index of land class i, and n is the sample size;

[0052] S22, sort the corresponding landscape ecological risk index under different dates according to the flow gradient under the corresponding date, and divide the sorted data into two groups, set one group of data as y1,y2,…,y h ,…,y n , these data are observed along the landscape gradient x1,x2,…,x n For continuous variables, the calculation formula of the bias is:

[0053]

[0054] Wherein, D is the bias of the whole data set, and μ is the mean of y h ;

[0055] S23, for each possible change point m, calculate the sum of the bias of the two subgroups, and compare it with the bias of the whole data set, the calculation formula of the bias reduction is:

[0056] Δ m = D-(D ≤m -D ＞m );

[0057] Wherein, D ≤m is the bias of the subgroup y1,y2,…,y m , D ＞m is the bias of the subgroup y m+1 ,y m+2 ,…,y n ;

[0058] The position t corresponding to the maximum bias reduction is identified as the change point:

[0059] t = argmax Δm ;

[0060] S24, extracting a plurality of random samples from the original data, calculating the probability of each position as a change point, combining the Bootstrap method to analyze the uncertainty, providing the confidence interval of the change point position by evaluating the uncertainty of the change point identification;

[0061] S25, according to the results of the non-parametric change point analysis, draw the frequency histogram and cumulative frequency curve, take the flow near the first steep rising point of the cumulative frequency curve as the initial flow of the initial growth period of vegetation, take the flow near the second steep rising point as the initial flow of the rapid growth period of vegetation, take the flow near the third steep rising point as the initial flow of the stable growth period of vegetation, and take the flow near the last steep rising point of the cumulative frequency curve as the peak flow of the stable growth period of vegetation.

[0062] Preferably, the landscape pattern index includes landscape loss degree index, landscape vulnerability index, landscape disturbance index, landscape fragmentation index, landscape separation index and landscape dominance index, and the calculation formula of the landscape pattern index is as follows:

[0063] R i =V i ×E i ;

[0064] E i =aC i +bN i +cF i ;

[0065]

[0066] Wherein, V i is the landscape vulnerability index of land class i, E i is the landscape disturbance index of land class i, C i is the landscape fragmentation index of land class i, N i is the landscape separation index of land class i, F i is the landscape dominance index of land class i, a is the weight of landscape fragmentation, b is the weight of landscape separation, c is the weight of landscape fractal dimension, and a+b+c=1, n f is the number of patches, A i is the area of land class i, A is the total area of the ecological risk area, Q f is the ratio of the number of sample plots where patch f appears to the total number of sample plots, M f is the ratio of the number of patches f to the total number of patches, L f is the ratio of the area of patch f to the total area of patch sample.

[0067] Preferably, the S3 comprises the following steps:

[0068] S31, import the daily flow data into the IHA software, analyze the five main flow components of the river, and generate the environmental flow process curve;

[0069] S32, select the vegetation growing season from April to October as the calculation period, and calculate the flow pulse occurrence time, flow pulse duration, flow pulse frequency and flow pulse rising and falling water rate respectively. The flow pulse refers to the process that the flow increases sharply to the peak value within a few hours to a few days, and then slowly falls back to the baseline level after maintaining a high flow state for a few hours to a few weeks, wherein the flow pulse duration is generally not less than three days;

[0070] S33, the rising duration is calculated by subtracting the initial flow of the vegetation growth stable period from the peak flow of the vegetation growth stable period and dividing by the rising rate, the falling duration is calculated by subtracting the initial flow of the vegetation growth stable period from the peak flow of the vegetation growth stable period and dividing by the falling rate, and the peak duration is calculated by subtracting the rising and falling time from the flow pulse duration. Namely:

[0071]

[0072] Peak duration = flow pulse duration - rising duration - falling duration;

[0073] Preferably, the S31 comprises the following steps:

[0074] Using a single parameter method, high flow events and low flow events are divided by setting a high flow threshold;

[0075] Based on the classification criteria of the minimum peak flow threshold of small flood and large flood, events reaching or exceeding these thresholds are classified as small flood or large flood;

[0076] 10% of the daily low flow is set as the extremely low flow threshold, and all flows lower than or equal to the threshold are marked as extremely low flow events.

[0077] Preferably, the S4 comprises the following steps:

[0078] S41, determine the landscape ecological water demand scheme according to each time node of the vegetation growing season:

[0079] In the early stage of vegetation growth, i.e. from April to mid-May, the ecological flow is set as the initial flow in the early stage of vegetation growth;

[0080] From mid-May to mid-June, the ecological flow is set as the initial flow in the rapid growth period of vegetation;

[0081] From mid-June to the day before the flow pulse occurs, the ecological flow is set as the initial flow in the stable growth period of vegetation;

[0082] The flow pulse occurs from January to mid-September, the ecological flow fluctuates between the initial flow in the stable period of vegetation growth and the peak flow in the stable period of vegetation growth, the fluctuation times are the flow pulse frequency, the duration of each fluctuation is the flow pulse duration, and the peak flow duration is the peak duration.

[0083] From mid-September to the end of October, the ecological flow is set to the initial flow in the stable period of vegetation growth.

[0084] S42, a landscape ecological water demand scheme according to S41 is drawn.

[0085] The present application has the following beneficial effects:

[0086] 1. The present application identifies the habitat characteristics and habitat requirements of plateau river wetland from the perspective of river wetland landscape pattern and ecological risk, and comprehensively uses various technical means around the type, fragmentation degree, connectivity and overall ecological health status of landscape unit.

[0087] 2. The present application closely links the ecological hydrological conditions with the vegetation phenology rhythm and the health status of the landscape ecological system, calculates the ecological hydrological indexes such as peak flow, duration, occurrence time, occurrence frequency and rising and falling water rate by the IHA method hydrology, and identifies the critical point and threshold of the change of flow change to the change of landscape ecological risk by using the non-parametric change point analysis, so as to intuitively and clearly analyze the influence of the change of ecological hydrological indexes on the landscape pattern and its ecological risk.

[0088] 3. The method provided by the present application can take the wetland of the ecological sensitive protection area as the key research object, quantitatively calculate the landscape ecological water demand of the plateau river wetland, and comprehensively judge the influence of the water conservancy construction on the downstream river wetland landscape ecology. DETAILED DESCRIPTION

[0089] Figure 1 It is a configuration method flowchart suitable for the landscape ecological water demand of plateau wetland of the embodiment of the present application;

[0090] Figure 2 It is a water system elevation map of the research area of the embodiment of the present application;

[0091] Figure 3 It is a random forest algorithm principle diagram based on a particle swarm optimization algorithm of the embodiment of the present application;

[0092] Figure 4 It is a landscape type classification result diagram of the embodiment of the present application;

[0093] Figure 5 It is a non-parametric change point analysis frequency cumulative frequency distribution diagram of the embodiment of the present application;

[0094] Figure 6 Fig. 1 is a schematic diagram of a vegetation growth season landscape ecological flow process line according to an embodiment of the present application. DETAILED DESCRIPTION

[0095] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below with reference to the drawings and embodiments.

[0096] An embodiment of the present application discloses a configuration method suitable for plateau wetland landscape ecological water demand, and the flow is as shown in Figure 1 The method comprises the following steps:

[0097] S1, remote sensing images are acquired, a random forest algorithm based on a particle swarm optimization algorithm is used for image classification, a landscape type classification map is generated, and the classification result is subjected to precision inspection and post-processing.

[0098] S11, in the embodiment, the Ako River wetland with unique ecological and hydrological characteristics is selected as the research area, and a water system elevation map of the Ako River basin is as shown in Figure 2 The L2A level cloud-free remote sensing image data of the area from 2019 to 2023 of Sentinel-2 is downloaded, and the sample points of the research area are extracted as a training set.

[0099] S12, spectral features, texture features, geometric features and context features are extracted from the remote sensing images to construct an initial feature set.

[0100] Spectral features are the most basic and commonly used features in remote sensing image classification, and reflect the reflection and radiation characteristics of ground objects in different bands. By utilizing the reflectivity difference between the red band and the near-infrared band, the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI) are used to distinguish vegetation and non-vegetation areas. By utilizing the reflectivity difference between the green band and the near-infrared band, the normalized water body index (NDWI) and the enhanced water body index (EWI) are used to distinguish water body and non-water body areas.

[0101] Texture features describe the spatial variation pattern of gray scale distribution in the image, and are important parameters reflecting the roughness and structural complexity of the surface of ground objects. Through the local binary pattern (LBP) analysis method, the gray scale values of the center pixel and its neighborhood pixels are compared to generate binary pattern codes, and statistics are constructed based on the gray level co-occurrence matrix as texture classification features. The image is subjected to multi-scale decomposition through wavelet transform, texture information at different scales is extracted, and the details and edge features in the image are captured, so as to effectively assist in distinguishing different landscape types with similar spectral features.

[0102] Geometric features describe the shape, size and structure of objects, which are important for distinguishing different types of objects. By calculating the area and perimeter of image objects, we can preliminarily distinguish different types of objects. By calculating the circularity, we can reflect the degree of object approaching to a circle. By calculating the aspect ratio, we can describe the length-width ratio of the object. By calculating the compactness, we can measure the smoothness of the object boundary.

[0103] Contextual features describe the spatial relationship and adjacency characteristics between image objects, which are important means to further improve the classification accuracy. By calculating the adjacency matrix between image objects, we can obtain the adjacency features and spatial layout information of objects, so as to more accurately identify the types of objects in complex scenes.

[0104] S13, use particle swarm optimization algorithm (Particle Swarm Optimization, PSO) to optimize the hyperparameters of random forest (Random Forest, RF), including the number of decision trees and the maximum depth, the method principle is as shown in Figure 3

[0105] S131, randomly initialize a group of particles in the hyperparameter space, each particle represents a possible parameter combination.

[0106] S132, update the speed and position of each particle according to the current speed, position, individual historical optimal position and global optimal position of the particle.

[0107] S133, use the performance index (such as classification accuracy) of random forest classifier on the training set as the fitness function, and calculate the fitness value of each particle.

[0108] S134, update the individual optimal position of each particle and the global optimal position of the whole particle group according to the fitness value.

[0109] S135, repeat S131-S134 until the preset stopping condition is met, such as reaching the maximum iteration round or the fitness value no longer has significant improvement.

[0110] S14, use the optimized random forest algorithm to evaluate the initial feature set, calculate the importance of the features, and construct a feature subset according to the importance of the features.

[0111] S141, use random forest algorithm to calculate the influence of each feature on performance during training, so as to obtain the importance of the feature.

[0112] S142, sort the features according to the importance of the features.

[0113] S143, select the top N most important features to construct the optimal feature subset. ​

[0114] S15, in the model training stage, a random forest classifier is trained on each classification node of the hierarchical classification tree using the selected feature subset to build a classification model.

[0115] S151, the optimal feature subset and the corresponding data subset are used to train a random forest classifier at each node.

[0116] S152, the random forest classifier is evaluated by calculating accuracy, recall rate, F1 score and other evaluation indicators to check whether its performance meets the requirements.

[0117] S153, the trained random forest classifier is connected to the corresponding node of the hierarchical classification tree to build a classification model.

[0118] S154, the classification model is optimized by adjusting the hierarchical structure, reselecting the feature subset and optimizing the random forest parameters to improve the overall performance of the classification model.

[0119] S16, in the classification and prediction stage, the image object to be classified is input into the hierarchical classification tree, and classification and prediction are performed layer by layer according to the classification rules and feature set to obtain the landscape type classification result. The specific landscape type classification result is shown in Figure 4 .

[0120] S17, in the interpretation process, for a small number of easily confused land classes, after visual interpretation and identification, high-resolution satellite images of Google Earth Pro are used for verification, and finally the plateau wetland landscape is divided into five categories according to the first-level classification standard (Land Use Classification GB / T21010-2017): beach (river heartland and beach), farmland (dry land), forest land (shrub forest and forest land), grassland (including natural grassland and artificial grassland), and water area (river with water). Through visual interpretation correction, a detailed landscape pattern distribution map is generated to show the distribution of the five landscape types.

[0121] S18, the classification result is evaluated by using the confusion matrix, and the accuracy of the classification result is comprehensively evaluated by using the indicators including user accuracy, producer accuracy, overall accuracy and Kappa coefficient.

[0122] User accuracy (User's Accuracy, UA) refers to the proportion of samples classified into a certain category that actually belong to that category, and the calculation formula is:

[0123]

[0124] Producer's Accuracy (PA) refers to the proportion of samples actually belonging to a certain class that are correctly classified, and the calculation formula is:

[0125]

[0126] Overall Accuracy (OA) refers to the proportion of correctly classified samples in the total number of samples, and the calculation formula is:

[0127]

[0128] where M ii is the diagonal element of the confusion matrix (i.e., the number of correctly classified samples), is the number of all samples predicted by the model as class i, and j is the other possible class, is the number of samples actually belonging to class i, and j is the other possible class, is the total number of samples.

[0129] The Kappa coefficient is a measure of the difference between the classification results and the random classification results, and the calculation formula is:

[0130]

[0131] where p0 is the overall accuracy, p e is the expected accuracy of random classification, k is the total number of classes, and N is the total number of samples.

[0132] S2, calculate the landscape ecological risk index by coupling the highland wetland landscape pattern index, and identify the mutation point of the landscape ecological risk index by non-parametric variable point analysis (nCPA) to determine the initial flow and peak flow in each vegetation growth period.

[0133] S21, calculate the landscape ecological risk index by further coupling the calculation of the landscape pattern index, and identify the high-risk area in the river wetland, the calculation formula of the landscape ecological risk index is:

[0134]

[0135] where A ki is the area of class i in the kth risk zone, A k is the area of the kth risk zone, R i is the landscape loss degree index of class i, and n is the sample size.

[0136] The landscape pattern index includes landscape loss degree index, landscape vulnerability index, landscape disturbance index, landscape fragmentation index, landscape separation index and landscape dominance index, and the calculation formula of the landscape pattern index is as follows:

[0137] R i =V i ×E i ;

[0138] E i =aC i +bN i +cF i ;

[0139]

[0140] wherein V i is the landscape vulnerability index of land class i, which is generally determined by expert scoring method, and the values are 5 for water area and beach, 4 for farmland, 3 for grassland, and 2 for forest land, E i is the landscape disturbance index of land class i, C i is the landscape fragmentation index of land class i, N i is the landscape isolation index of land class i, F i is the landscape dominance index of land class i, a is the weight of the landscape fragmentation, b is the weight of the landscape isolation, c is the weight of the landscape fractal dimension, and a+b+c=1, in the embodiment, a is 0.5, b is 0.3, and c is 0.2.n f is the number of patches, A i is the area of land class i, A is the total area of the ecological risk zone, Q f is the ratio of the number of sample plots in which patch f appears to the total number of sample plots, M f is the ratio of the number of patches in which patch f appears to the total number of patches, and L f is the ratio of the area of patch f to the total area of the sample plots.

[0141] S22, sort the corresponding landscape ecological risk indexes on different dates according to the flow gradient on the corresponding dates, and divide the sorted data into two groups, and set one group of data as y1, y2, …, y h , …, y n , which are observed along the landscape gradient x1, x2, …, x n For continuous variables, the calculation formula of the bias is:

[0142]

[0143] wherein D is the bias of the entire data set, and μ is the mean of y h .

[0144] S23, for each possible change point m, calculate the sum of the bias of the two subgroups, and compare it with the bias of the entire data set, and the calculation formula of the bias reduction amount is:

[0145] Δ m= D - (D ≤m - D ＞m );

[0146] where D ≤m is the deviation of the subgroup y1, y2, …, y m , and D ＞m is the deviation of the subgroup y m+1 , y m+2 , …, y n ;

[0147] The position t corresponding to the maximum deviation reduction is identified as the change point:

[0148] t = argmax D m ;

[0149] S24, extract a plurality of random samples from the original data, calculate the probability of each position as a change point, combine the Bootstrap method for uncertainty analysis, evaluate the uncertainty of the change point identification, and provide the confidence interval of the change point position.

[0150] Through nCPA analysis, first sort the landscape ecological risk index and flow data, and divide them into different intervals. By calculating the deviation in each interval, the point where the landscape ecological risk index changes significantly at the flow level is identified.

[0151] S25, according to the results of non-parametric variable point analysis, draw a frequency histogram and a cumulative frequency curve, as shown in Figure 5 . The flow near the first steep rising point of the cumulative frequency curve (about 3.00 m 3 / s) is taken as the initial flow in the early stage of vegetation growth (April 1 to mid-May), marking the beginning of significant changes in the landscape ecosystem. The flow near the second steep rising point (about 5.41 m 3 / s) is taken as the initial flow in the rapid growth period of vegetation (mid-May to mid-June), at which time the landscape ecological risk index decreases significantly. The flow near the third steep rising point (about 15.14 m 3 / s) is taken as the initial flow in the stable growth period of vegetation (mid-June to mid-September). The flow near the last steep rising point of the cumulative frequency curve (about 38.04 m 3 / s) is taken as the peak flow in the stable growth period of vegetation (mid-June to mid-September). At this time, the landscape ecosystem is almost in the optimal state, and the ecological risk index is at the lowest, meaning that the health and stability of the ecosystem are maximally supported at this flow level.

[0152] S3, Selecting hydrological elements with ecological significance, such as flow, frequency, occurrence time, duration, and rate of change, the ecological hydrological index is used to quantitatively analyze the change of wetland hydrological regime.

[0153] S31, Importing daily flow data into IHA software, analyzing five main flow components of the river, namely low flow, very low flow, high flow pulse, small flood and large flood, and generating environmental flow process curve.

[0154] In this embodiment, the ecological hydrological index is calculated by IHA software. First, the initial separation method is used to identify high flow events and low flow events, and the single parameter method is used to divide high flow events and low flow events by setting high flow threshold (75% of daily flow in this embodiment). In order to reflect the ecological demand more accurately, four parameter method is used for calibration, including high flow threshold, low flow threshold, high flow start rate threshold and high flow end rate threshold, to control the start and end of high flow event. Then, high flow event is further classified into high flow pulse, small flood and large flood. The classification standard is based on the minimum peak flow threshold of small flood and large flood, and the events reaching or exceeding these thresholds are classified as small flood or large flood. Finally, the very low flow event is identified by setting the very low flow threshold. 10% of daily low flow is set as the very low flow threshold, and all flows lower than or equal to the threshold are marked as very low flow events. Parameter calibration is an important step in EFC calculation, which ensures that the algorithm accurately reflects the ecological conditions and needs of the river. Calibration is usually divided into two stages, first, correct separation of high flow and low flow, and second, adjustment of high flow and very low flow event classification.

[0155] S32, Selecting 4-10 months of vegetation growing season as calculation period, respectively calculating flow pulse occurrence time, flow pulse duration, flow pulse frequency and flow pulse rising and falling water rate. Flow pulse refers to the process that flow increases sharply to peak within hours to days, and maintains high flow state for hours to weeks before slowly falling back to baseline level. Among them, the flow pulse duration is generally not less than three days;

[0156] S33, The rising duration is calculated by subtracting the initial flow of the vegetation growing stable period from the peak flow of the vegetation growing stable period and dividing by the rising rate, the falling duration is calculated by subtracting the initial flow of the vegetation growing stable period from the peak flow of the vegetation growing stable period and dividing by the falling rate, and the peak duration of flow pulse is calculated by subtracting the rising and falling time from the flow pulse duration. That is:

[0157]

[0158] Peak duration = flow pulse duration - rising duration - falling duration;

[0159] The vegetation growth season landscape ecological hydrological index values shown in Table 1 are obtained through S31-S33.

[0160] Table 1 Vegetation growth season landscape ecological hydrological index

[0161]

[0162]

[0163] S4, coupling ecological hydrological processes, determining the landscape ecological water demand scheme at each time node of the vegetation growth season, and drawing the landscape ecological water demand process.

[0164] S41, performing ecological hydrological index analysis on the measured flow data at the research area, establishing a flow pulse process, determining the landscape ecological water demand scheme according to each time node of the vegetation growth season, and drawing the landscape ecological water demand process as shown in Figure 6 .

[0165] In the early stage of vegetation growth, i.e. from April to mid-May, the ecological flow is set to the initial flow in the early stage of vegetation growth, i.e. 3m 3 / s, aiming to maintain the basic functions of river and wetland ecosystems and ensure the minimum water demand required for vegetation germination and early growth. The monthly average flow is 7.72m 3 / s, and the ecological flow accounts for 38.8%.

[0166] From mid-May to mid-June, the ecological flow is set to the initial flow in the rapid growth period of vegetation, i.e. 5.41m 3 / s, to meet the needs of accelerating growth and ensure habitat connectivity and the maintenance of biodiversity. Although the monthly average flow increases to 18.78m 3 / s, the proportion of ecological flow decreases to 22.60%, reflecting that the increase in water volume during this stage is greater than the ecological needs.

[0167] From mid-June to the day before the flow pulse occurs, the vegetation growth enters the stable period, and the ecological flow is set to the initial flow in the stable period of vegetation growth, i.e. 15.14m 3 / s, to promote the improvement of landscape ecology. At this time, the monthly average flow is 47.50m 3 / s, but the proportion of ecological flow decreases to 21.63%.

[0168] The flow pulse occurs from May to mid-September, and the ecological flow fluctuates between the initial flow in the stable growth period of vegetation and the peak flow in the stable growth period of vegetation, with the fluctuation frequency being the flow pulse frequency, the fluctuation duration being the flow pulse duration, and the peak flow duration being the peak duration. By implementing 6 flow pulses, the landscape morphology and ecological pattern are artificially changed, the floodplain is promoted to renew, and the structure and function of the riparian zone are profoundly affected. The flow pulse design in this stage reflects the consideration of flood events and their ecological effects.

[0169] From mid-September to the end of October, the ecological flow is set to the initial flow in the stable growth period of vegetation, i.e., 15.14 m 3 / s, to meet the needs of vegetation entering the late growth period and dormancy period. Although the monthly average flow is still as high as 31.05 m 3 / s, the ecological flow ratio rises to 48.77%.

[0170] In summary, the dynamic adjustment strategy of ecological flow aims to reflect the changes in the functional needs of the landscape ecology in different growth stages of vegetation.

[0171] S42, draw a landscape ecological water demand process according to the landscape ecological water demand scheme of S41.

[0172] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for configuring water for plateau wetland landscape ecology, characterized in that: The following steps are involved: S1. Obtain remote sensing images, use the random forest algorithm based on the particle swarm optimization algorithm to perform image classification, generate landscape type classification maps, and perform accuracy testing and post-processing on the classification results; S2. Calculate the landscape ecological risk index by coupling the plateau wetland landscape pattern index, use non-parametric change point analysis to identify the mutation points of flow to the landscape ecological risk index, and determine the initial flow and peak flow of each vegetation growth period; S3. Select hydrological elements with ecological significance and use eco-hydrological indicators to quantitatively analyze changes in wetland hydrological conditions; S4. Couple the eco-hydrological process, determine the landscape ecological water demand plan at each time point in the vegetation growing season, and draw the landscape ecological water demand process.

2. The method for configuring water for plateau wetland landscape ecology according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Download Sentinel-2 L2A cloud-free remote sensing image data and extract sample points in the study area as a training set. S12, extracting spectral features, texture features, geometric features and context features from remote sensing images to construct an initial feature set; S13. Use particle swarm optimization algorithm to optimize the hyperparameters of random forest; S14. Use the optimized random forest algorithm to evaluate the initial feature set, calculate the importance of the features, and sort the features according to their importance to construct feature subsets; S15. At each classification node of the hierarchical classification tree, a random forest classifier is trained using the selected feature subset to construct a classification model; S16, inputting the image object to be classified into a hierarchical classification tree, classifying and predicting layer by layer according to the classification rules and feature set, and obtaining the landscape type classification result; S17. During the interpretation process, easily confused land types were identified through visual interpretation and verified with the help of high-resolution satellite images from Google Earth Pro. Ultimately, the plateau wetland landscape was divided into five categories: beach, cultivated land, woodland, grassland, and water, and a detailed landscape pattern distribution map was generated. S18. Use the confusion matrix to evaluate the accuracy of the classification results, and comprehensively evaluate the accuracy of the classification results through indicators including user accuracy, producer accuracy, overall accuracy and Kappa coefficient.

3. The method for configuring water for plateau wetland landscape ecology according to claim 2, characterized in that: The S13 includes the following steps: S131. Randomly initialize a set of particles in the hyperparameter space, where each particle represents a possible parameter combination; S132. Update the speed and position of each particle based on the particle's current speed, position, individual historical optimal position, and global optimal position; S133, using the performance index of the random forest classifier on the training set as the fitness function to calculate the fitness value of each particle; S134, updating the individual optimal position of each particle and the global optimal position of the entire particle swarm according to the fitness value; S135. Repeat S131-S134 until a preset stop condition is met.

4. The method for configuring water for plateau wetland landscape ecology according to claim 3, characterized in that: The S14 includes the following steps: S141. Use the random forest algorithm to calculate the impact of each feature on performance during training, thereby deriving the importance of the feature. S142. Sort features according to their importance; S143. Select the top N most important features to construct the optimal feature subset.

5. The method for configuring water for plateau wetland landscape ecology according to claim 4, characterized in that: The S15 comprises the following steps: S151. Use the optimal feature subset and the corresponding dataset subset to train a random forest classifier on each node; S152. Evaluate the random forest classifier by calculating evaluation indicators to check whether its performance meets the requirements; S153, connecting the trained random forest classifier to the corresponding node of the hierarchical classification tree to build a classification model; S154. The classification model is optimized by adjusting the hierarchical structure, reselecting feature subsets, and optimizing random forest parameters to improve the overall performance of the classification model.

6. The method for configuring water for plateau wetland landscape ecology according to claim 5, characterized in that: The S2 comprises the following steps: S21. By calculating the landscape pattern index, we further couple the calculation of the landscape ecological risk index to identify high-risk areas in river wetlands. The landscape ecological risk index calculation formula is: Among them, A ki is the area of ​​land type i in the k-th risk cell, A k is the area of ​​the kth risk cell, R i is the landscape loss index of land type i, and n is the sample size; S22. Sort the landscape ecological risk indexes corresponding to different dates according to the flow gradients on the corresponding dates, and divide the sorted data into two groups. Let one group of data be y1, y2, ..., y h ,...,y n , these data are along the landscape gradient x1, x2, ..., x n For continuous variables, the deviation is calculated as: Where D is the deviation of the entire data set, μ is y h The mean of S23. For each possible change point m, calculate the sum of the deviations of the two subgroups and compare it with the deviation of the entire data set. The formula for calculating the deviation reduction is: Δ m =D-(D ≤m -D >m ); Where D≤m is the subgroup y1, y2, ..., y m Deviation, D>m is the subgroup y m+1 ,y m+2 ,...,y n Deviation; The position t corresponding to the maximum deviation reduction is identified as the change point: t=argmaxΔ m ; S24. Extract multiple random samples from the original data, calculate the probability of each position being a change point, and perform uncertainty analysis in combination with the Bootstrap method. By evaluating the uncertainty of change point identification, a confidence interval for the change point location is provided. S25. Based on the results of non-parametric change point analysis, draw a frequency histogram and a cumulative frequency curve. The flow rate near the first steep rising point of the cumulative frequency curve is taken as the initial flow rate in the early stage of vegetation growth, the flow rate near the second steep rising point is taken as the initial flow rate in the rapid growth period of vegetation, the flow rate near the third steep rising point is taken as the initial flow rate in the stable growth period of vegetation, and the flow rate near the last steep rising point of the cumulative frequency curve is taken as the peak flow rate in the stable growth period of vegetation.

7. The method for configuring water for plateau wetland landscape ecology according to claim 6, characterized in that: The landscape pattern index includes landscape loss index, landscape vulnerability index, landscape disturbance index, landscape fragmentation index, landscape separation index and landscape dominance index. The calculation formula of landscape pattern index is as follows: R i =V i ×E i ; E i =aC i +bN i + cF i ; Among them, V i is the landscape vulnerability index of land type i, E i is the landscape disturbance index of land type i, C i is the landscape fragmentation index of land type i, N i is the landscape separation index of land type i, F i is the landscape dominance index of land type i, a is the weight of landscape fragmentation, b is the weight of landscape separation, c is the weight of landscape fractal dimension, and a+b+c=1, n f is the number of patches, Ai is the area of ​​land type i, A is the total area of ​​ecological risk zone, Q f is the ratio of the number of sample areas where patch f appears to the total number of sample areas, M f is the ratio of the number of plaques f to the total number of plaques, L f It is the ratio of the area of ​​patch f to the total area of ​​the patch sample area.

8. The method for configuring water for plateau wetland landscape ecology according to claim 7, characterized in that: The S3 includes the following steps: S31. Import daily flow data into IHA software, analyze the five main flow components of the river, and generate environmental flow process curves; S32. Select April to October during the vegetation growing season as the calculation period, and calculate the flow pulse occurrence time, flow pulse duration, flow pulse frequency, and flow pulse rise and fall rate. The flow pulse duration is no less than three days. S33. The duration of water rise is calculated by subtracting the initial flow in the stable vegetation growth period from the peak flow in the stable vegetation growth period and dividing it by the water rise rate. The duration of water retreat is calculated by subtracting the initial flow in the stable vegetation growth period from the peak flow in the stable vegetation growth period and dividing it by the water retreat rate. The duration of flow pulse peak is calculated by subtracting the water rise and fall time from the flow pulse duration.

9. The method for configuring water for plateau wetland landscape ecology according to claim 8, characterized in that: The S31 includes the following steps: A single-parameter method is used to classify high-flow events and low-flow events by setting a high-flow threshold; Based on the classification criteria for minimum peak flow thresholds for minor and major floods, events that meet or exceed these thresholds are classified as minor or major floods; 10% of the daily low flow rate is set as the extremely low flow threshold, and all flows below or equal to this threshold are marked as extremely low flow events.

10. The method for configuring water for plateau wetland landscape ecology according to claim 9, characterized in that: The S4 comprises the following steps: S41. Determine the landscape ecological water demand plan based on the various time points during the vegetation growth season: In the early stage of vegetation growth, i.e., from April to mid-May, the ecological flow was set as the initial flow at the beginning of vegetation growth; From mid-May to mid-June, the ecological flow was set as the initial flow during the rapid vegetation growth period; From mid-June to the day before the flow pulse, the ecological flow was set as the initial flow during the stable vegetation growth period; From the day of the flow pulse to mid-September, the ecological flow fluctuated between the initial flow during the vegetation growth stability period and the peak flow during the vegetation growth stability period. The number of fluctuations was the flow pulse frequency, the duration of each fluctuation was the flow pulse duration, and the duration of the peak flow was the peak duration. From mid-September to the end of October, the ecological flow was set as the initial flow during the stable vegetation growth period; S42. Draw the landscape ecological water demand process based on the landscape ecological water demand plan in S41.

Images