Grassland degradation degree monitoring model and method based on remote sensing technology

Abstract

The application discloses a grassland degradation degree monitoring model and method based on remote sensing technology, and belongs to the technical field of ecological environment monitoring. The method comprises the following steps: step 1, constructing a multi-scale remote sensing data set consistent in time and space, and combining ground measured data to establish a high-precision true value library of grassland types and degradation degrees; step 2, constructing a remote sensing index system suitable for different spatial scales; step 3, identifying key scale conversion nodes of grassland degradation driven by climate and human factors; step 4, realizing grassland degradation degree grading from a pixel scale to a landscape scale, and quantitatively analyzing the spatial distribution, area change and spatio-temporal evolution trend of each degradation grade; step 5, realizing reliable conversion and up-down transmission of analysis results of different scales to generate comprehensive grassland degradation monitoring results; and step 6, automatically generating differentiated management strategies and outputting a grassland degradation grade distribution map, a driving force analysis map, a risk early warning map and a partition management suggestion map.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological environment monitoring, more specifically, to a grassland degradation degree monitoring model and method based on remote sensing technology. BACKGROUND

[0002] Grassland degradation refers to the phenomenon that grassland ecosystems are affected by natural factors and human activities, leading to a decrease in vegetation coverage, deterioration of soil quality, and decline in ecological functions. Grassland degradation not only affects the stability of the ecological environment, but also leads to soil erosion, reduction of biodiversity, and decline in agricultural and pastoral productivity. With the increasing severity of grassland degradation, how to achieve real-time monitoring and accurate assessment of the grassland degradation process has become a key problem that needs to be solved globally.

[0003] Traditional methods of grassland degradation monitoring usually rely on ground surveys and manual sampling. These methods are not only time-consuming and labor-intensive, but also have certain limitations, making it difficult to cover a large area and unable to reflect the dynamic changes of grassland degradation in real time. With the rapid development of remote sensing technology, remote sensing-based grassland monitoring methods have gradually become the main means of grassland degradation monitoring due to their ability to quickly obtain efficient data over a large area. However, existing remote sensing monitoring methods still face several challenges, including how to accurately assess the degree of grassland degradation at multiple scales and time scales, how to integrate remote sensing data from different sources, how to effectively identify the different effects of climate and human factors on grassland degradation, and how to provide precise and operational support for management decisions.

[0004] Although remote sensing technology has made some progress in grassland degradation monitoring, existing methods mostly focus on single-scale or single-index monitoring, lacking comprehensive analysis of multi-dimensional characteristics, especially in the aspects of multi-scale data fusion and degradation driving force analysis with spatio-temporal consistency. Meanwhile, existing research on the application of grassland degradation monitoring results often lacks scientific and reasonable management strategies and measures, making it difficult to truly promote grassland protection and restoration work.

[0005] In summary, how to achieve accurate monitoring, comprehensive analysis, and efficient management strategy generation of grassland degradation through remote sensing technology has become a technical problem that needs to be solved. SUMMARY

[0006] In order to overcome the series of defects existing in the prior art, the purpose of the present application is to provide a grassland degradation degree monitoring method based on remote sensing technology, which includes the following steps:

[0007] Step 1, construct a spatio-temporally consistent multi-scale remote sensing dataset, and combine ground measured data to establish a high-precision true value library of grassland types and degradation degrees;

[0008] Step 2: Automatically screen remote sensing indicators most sensitive to grassland degradation, and build remote sensing indicator system adaptive to different spatial scales;

[0009] Step 3: Identify key scale transition nodes driven by climate and human factors for grassland degradation;

[0010] Step 4: Realize grassland degradation grading from pixel scale to landscape scale, and quantitatively analyze spatial distribution, area change and spatio-temporal evolution trend of each degradation grade;

[0011] Step 5: Realize reliable conversion and up-down transmission of analysis results of different scales, and generate comprehensive grassland degradation monitoring results;

[0012] Step 6: Automatically generate differentiated management strategies, and output grassland degradation grade distribution map, driving force analysis map, risk warning map and partition management suggestion map.

[0013] Further, step 2 includes the following steps:

[0014] Based on the corrected multi-source remote sensing data, calculate vegetation index, water index and soil index, and build a comprehensive index set that can fully represent the structural and functional characteristics of grassland ecosystem;

[0015] For each index, calculate its statistical characteristics in time and space dimensions, build a multi-dimensional feature space, and accurately depict the dynamic changes of grassland ecosystem in different degradation stages;

[0016] On the three spatial scales of pixel, patch and landscape, calculate the information gain value and mutual information value of each index for grassland degradation grading, quantitatively evaluate the classification contribution of each index at different scales, and select the candidate index set with significant and stable contribution through information stability analysis;

[0017] Evaluate the response sensitivity of each index to grassland degradation gradient, calculate the sensitivity coefficient and discrimination index, and analyze the performance differences of the index in different ecological regions, grassland types and degradation degree intervals, combined with the true value library data, to identify the feature index with the most discrimination ability at each scale;

[0018] Automatically screen the optimal index combination at different spatial scales to ensure classification accuracy while minimizing redundancy between indexes, and consider the economy and practicality of data acquisition to form an adaptive index combination scheme adaptive to different scales and grassland types;

[0019] Establish cross-scale index transmission relationship and conversion rules, evaluate the classification performance and stability of each scale index system based on independent verification data set, and finally build a grassland monitoring index system with scale adaptability, regional pertinence and degradation sensitivity.

[0020] Further, for each index, calculate its statistical characteristics in the time dimension and the spatial dimension, construct a multi-dimensional feature space, and accurately depict the dynamic changes of the grassland ecosystem in different degradation stages, including the following steps:

[0021] For each index, calculate the time mean μ t , the time variance σ t 2 , the change rate V t , and the autocorrelation R t to capture the stability, volatility, and trend information of the grassland ecosystem at different times;

[0022] At each time, calculate the spatial mean μ s , the spatial variance σ s 2 , and the spatial autocorrelation I s for all index data to quantify the balance, dispersion, and adjacency of the ecosystem in the spatial distribution;

[0023] Combine the statistical characteristics extracted in the time and spatial dimensions into a multi-dimensional feature vector F = [μ t , σ t 2 , V t , R t , μ s , σ s 2 , I s ] to form a feature space that comprehensively depicts the dynamic changes of grassland degradation;

[0024] By comparing the feature vectors of different periods or regions and calculating the ecosystem dynamic change index D, accurate determination and dynamic monitoring of the degradation stage of the grassland ecosystem can be achieved, and the formula is: where μ t (t) is the index mean in the time dimension, representing the mean of the grassland degradation index calculated at time t; μ s (s) is the index mean in the spatial dimension, representing the mean of the grassland degradation index calculated at spatial unit s; μ t (t+1) is the index mean in the time dimension, representing the mean of the grassland degradation index calculated at time t+1; and μ s (s+1) is the index mean in the spatial dimension, representing the mean of the grassland degradation index calculated at spatial unit s+1.

[0025] Further, at the pixel, patch, and landscape spatial scales, calculate the information gain value and mutual information value of each index for grassland degradation classification, quantitatively evaluate the classification contribution of each index at different scales, and select a candidate index set with significant and stable contribution through information stability analysis, including the following steps:

[0026] According to the remote sensing data of the target area, pixels, patches and landscapes are divided into three spatial scales;

[0027] At each spatial scale, the corresponding grassland degradation classification label and each remote sensing index data are collected to ensure that the data set at each scale contains sufficient sample size and spatial information;

[0028] For each spatial scale, the information gain formula IG(T,A) = H(T)-H(T|A) is used to calculate the information gain value of each remote sensing index A to the grassland degradation classification T, wherein IG(T,A) represents the information gain of remote sensing index A to the grassland degradation classification T, that is, the degree of uncertainty reduction of the grassland degradation classification under the condition of knowing A; H(T) represents the entropy of the grassland degradation classification T, which measures the overall uncertainty of T; H(T|A) represents the conditional entropy of the grassland degradation classification T given the remote sensing index A, that is, the remaining uncertainty of T under the condition of knowing A;

[0029] The mutual information formula MI(A,T) = H(T)+H(A)-H(T,A) is used to calculate the mutual information value between the remote sensing index and the grassland degradation classification at different spatial scales, wherein MI(A,T) represents the mutual information between the remote sensing index A and the grassland degradation classification T; H(A) represents the entropy of the remote sensing index A; H(T,A) represents the joint entropy of the grassland degradation classification T and the remote sensing index A;

[0030] The information gain value and the mutual information value are used to quantitatively evaluate the classification contribution of each remote sensing index at different spatial scales;

[0031] The standard deviation of the information gain or mutual information value of the remote sensing index at different spatial scales is calculated, and the formula is used: to evaluate the stability of information, wherein μ IG represents the mean value of the information gain value calculated at different spatial scales, which is used to evaluate the classification contribution of the index as a whole; σ IG represents the standard deviation of the information gain value calculated at different spatial scales, which is used to measure the stability of the classification contribution of the index at each scale;

[0032] According to the information stability analysis result, the remote sensing index set which is stable and has significant contribution at different scales is selected.

[0033] Further, step 3 includes the following steps:

[0034] Integrate climate factor and human factor data to establish a spatiotemporally consistent grassland degradation potential driving force database to ensure that each driving factor data matches the remote sensing monitoring data in terms of time and spatial resolution;

[0035] With grassland degradation index as response variable, climate and human factors as predictor variable, a multi-level Bayesian network model is constructed, wherein: the first level represents the response at pixel scale, the second level represents the response at patch scale, and the third level represents the response at landscape scale, and each level is connected through conditional probability relationship;

[0036] The main effect, interaction effect and contribution rate of climate and human factors at different scales are calculated, and the independent contribution of each factor is stripped by using partial correlation analysis and variance decomposition technique.

[0037] Combined with information criterion and Bayesian factor comparison, the spatial scale conversion threshold of dominant effect of climate and human factors is identified, and the optimal scale segmentation point is determined by using entropy maximization criterion, and a scale-driving force response curve is constructed to identify the key scale node.

[0038] The relative contribution rate of each driving factor to grassland degradation at different scales and its spatial differentiation characteristics are calculated, and a scale-driving force contribution rate map is generated to realize the accurate quantification and spatial visualization of the influence of climate and human factors.

[0039] Further, the main effect, interaction effect and contribution rate of climate and human factors at different scales are calculated, and the independent contribution of each factor is stripped by using partial correlation analysis and variance decomposition technique, including the following steps:

[0040] Using a multiple linear regression model, set the grassland degradation degree as the dependent variable, and the climate factors and human factors as the independent variables, the main effect coefficient is obtained by regression analysis, and the contribution rate of each main effect to the total variance is calculated;

[0041] The interaction term of climate factors and human factors is added to the regression model, the interaction effect coefficient is calculated, and the contribution rate of the interaction effect is calculated by variance analysis to measure the influence of the interaction of climate and human factors on grassland degradation;

[0042] The partial correlation coefficient and respectively assess the independent contribution of climate and human factors to grassland degradation under the control of the influence of the other factor, wherein, represents the partial correlation coefficient between variable X1 and dependent variable Y after controlling variable X2; represents the partial correlation coefficient between variable X2 and dependent variable Y after controlling variable X1;

[0043] The total variance is decomposed into the variance of climate factors, human factors, interaction effect and error term by using variance decomposition method;

[0044] By calculating the variance contribution rate of each part, the contribution degree of each factor to the variation of grassland degradation is determined.

[0045] Further, step 4 comprises the following steps:

[0046] At the pixel scale, local texture features are extracted using an improved convolutional neural network; at the patch scale, an attention mechanism is integrated to enhance the expression of key regional features; and at the landscape scale, a graph convolution network is introduced to capture the topological relationship between landscape units.

[0047] The spatial context information is used to optimize the classification boundary, realizing high-precision grassland degradation classification at a resolution of 1-30 meters.

[0048] The high-resolution remote sensing image is multi-scale segmented to generate homogeneous patches with ecological significance, realizing degradation classification at the functional unit level of grassland, and analyzing the impact of grassland fragmentation and connectivity changes on ecological function.

[0049] The macro pattern and functional changes of the grassland ecosystem are captured, the spatial aggregation and diffusion trend of the degraded area are analyzed, and the integrity of the ecosystem is evaluated.

[0050] Based on the multi-period classification results, a degradation transition matrix is constructed to quantitatively calculate the area proportion, spatial distribution characteristics and temporal evolution law of each degradation level at different scales.

[0051] Further, step 5 comprises the following steps:

[0052] A conversion rule library is established among the three scales of pixels, patches and landscapes. For discrete classification results, the dominant type method and weighted voting method are used for upscaling aggregation; for continuous degradation index, the regional average method and area weighting method are used for scale conversion; at the same time, a nonlinear conversion function considering spatial heterogeneity is designed to capture the mutation effect of degradation characteristics in the scale conversion process.

[0053] The global information at a high level scale is used as prior knowledge to constrain the local detail expression at a lower level scale, and the high-precision information at a lower level scale is used as evidence to update the overall cognition at a higher level scale, constructing a bidirectional information flow mechanism.

[0054] The entropy, variance and confidence interval of each spatial unit at different scales are calculated to identify high-uncertainty areas and analyze the main factors causing uncertainty.

[0055] Grassland types and degradation processes that are highly sensitive to a particular scale are identified, the scale adaptability index is calculated, the applicability of monitoring results at each scale is quantitatively evaluated, and an optimal scale selection decision tree is constructed.

[0056] The monitoring results at each scale are fused according to the reliability weight to generate a unified precision standard and a comprehensive grassland degradation monitoring result with complete spatial coverage.

[0057] Further, step 6 includes the following steps:

[0058] Based on the grassland degradation level classification results and spatial statistical analysis, a high-precision degradation level distribution map is generated, and the spatial distribution characteristics of different degradation degree regions are visually displayed through a graded color scale;

[0059] A spatial contribution rate model of climate factors and human factors is constructed to generate a driving force analysis map, and a double-factor color coding method is used to visually present the spatial heterogeneity of the dominant factors;

[0060] Combined with historical data sequences and current monitoring results, a grassland degradation risk warning map is generated to identify potential rapid degradation areas and ecological fragile points, and according to the risk level, the corresponding warning threshold is set;

[0061] According to the grassland degradation degree, driving factors and ecological sensitivity, management units are divided, and a partition management suggestion map is generated to propose targeted management measures to achieve the efficiency and pertinence of grassland protection and restoration.

[0062] The grassland degradation degree monitoring model based on remote sensing technology includes the following modules:

[0063] The data acquisition and preprocessing module is responsible for integrating multi-source remote sensing data and ground measured data, constructing a multi-scale data set consistent in time and space, and preprocessing to ensure the accuracy and consistency of the data;

[0064] The degradation degree evaluation module calculates relevant indicators and performs multi-dimensional feature analysis based on the preprocessed remote sensing data, and selects the optimal index combination;

[0065] The driving force analysis module quantifies the main effect, interaction effect and contribution rate of climate and human factors on grassland degradation, and identifies key scale nodes;

[0066] The result display and management module displays the grassland degradation level distribution, driving force analysis, risk warning and management suggestion in the form of intuitive charts, and performs management;

[0067] The multi-scale analysis and conversion module establishes conversion rules between different spatial scales, realizes multi-scale segmentation and feature extraction, constructs a two-way information flow mechanism, and ensures reliable conversion and fusion of different scale analysis results;

[0068] The management strategy generation module automatically generates differentiated management strategies according to the grassland degradation monitoring results and related analysis, proposes targeted management measures, and realizes the efficiency and pertinence of grassland protection and restoration.

[0069] Compared with the prior art, the present application has the following beneficial effects:

[0070] The application realizes dynamic monitoring of grassland degradation degree, degradation driving force analysis and differentiated management strategy generation by constructing a multi-scale remote sensing dataset and comprehensively analyzing climate and human factors, combined with a high-precision true value library, and provides an accurate, visual and operable grassland degradation evaluation and management solution. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 A flowchart of a grassland degradation degree monitoring method based on remote sensing technology disclosed in an embodiment of the application.

[0072] Figure 2 A structural diagram of a grassland degradation degree monitoring model based on remote sensing technology disclosed in an embodiment of the application. DETAILED DESCRIPTION

[0073] For the purpose, technical scheme and advantages of the implementation of the application, the technical scheme in the embodiment of the application will be described in more detail below in combination with the drawings in the embodiment of the application. In the drawings, the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The described embodiments are part of the embodiments of the application, not all of the embodiments.

[0074] Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the application.

[0075] The embodiments described below and the directional words are exemplary and are intended to explain the application, and cannot be understood as a limitation on the application.

[0076] As shown in Figure 1 The grassland degradation degree monitoring method based on remote sensing technology comprises the following steps:

[0077] Step 1, constructing a multi-scale remote sensing dataset consistent in time and space, and combining ground measured data to establish a high-precision true value library of grassland types and degradation degree;

[0078] Step 2, automatically selecting remote sensing indexes most sensitive to grassland degradation, and constructing a remote sensing index system suitable for different spatial scales;

[0079] Step 3, identifying key scale conversion nodes of grassland degradation driven by climate and human factors;

[0080] Step 4, realizing grassland degradation degree classification from pixel scale to landscape scale, and quantitatively analyzing the spatial distribution, area change and spatio-temporal evolution trend of each degradation grade;

[0081] Step 5: Realize reliable conversion and up-down transmission of analysis results of different scales, and generate comprehensive grassland degradation monitoring results;

[0082] Step 6: Automatically generate differentiated management strategies, and output grassland degradation level distribution map, driving force analysis map, risk warning map, and partition management suggestion map.

[0083] In summary, the grassland degradation degree monitoring method based on remote sensing technology effectively improves the accuracy and spatio-temporal adaptability of grassland degradation monitoring through a series of accurate data processing and analysis steps. First, a spatio-temporally consistent multi-scale remote sensing dataset is constructed, and combined with ground measured data, a high-precision grassland type and degradation degree true value library is established, providing reliable basic data for subsequent analysis. Then, the remote sensing index system suitable for different spatial scales is constructed by automatically selecting the remote sensing index that best reflects the grassland degradation, which ensures the multidimensionality and accuracy of the monitoring results. In identifying the key driving factors of grassland degradation, this method pays special attention to the key scale conversion nodes under the action of climate and human factors, thereby revealing the complex spatio-temporal dynamics of grassland degradation. Then, through the classification of grassland degradation degree from pixel scale to landscape scale, the accurate definition of degradation level is realized, and through the quantitative analysis of its spatial distribution, area change and spatio-temporal evolution trend, the depth and breadth of the monitoring results are further enhanced. In addition, this method realizes the reliable conversion and up-down transmission of analysis results between different scales, so that the monitoring data can effectively communicate between different spatial levels, and finally generate comprehensive grassland degradation monitoring results. On this basis, this method can automatically generate differentiated grassland management strategies, and output grassland degradation level distribution map, driving force analysis map, risk warning map, and partition management suggestion map, providing practical decision support for grassland protection and restoration.

[0084] Further, step 1 includes the following steps:

[0085] High-resolution optical satellite images are used for fine grassland mapping, medium and low-resolution optical satellite images are used for long-term dynamic monitoring, and microwave remote sensing data are used for cloud-penetrating observation of grassland biomass and moisture conditions, and all images are subjected to strip removal, broken row repair and denoising processing;

[0086] Based on high-precision control point network, geometric precision correction is performed on remote sensing images from different sources to ensure pixel-level registration between different data sources;

[0087] Atmospheric correction is performed on optical remote sensing images to eliminate atmospheric scattering and absorption effects, and terrain correction is performed on mountain grassland areas to remove terrain shadows and slope effects;

[0088] The data cubes with unified time steps and multi-spatial scales are constructed by combining high spatial resolution and high temporal resolution data, and the spectral consistency between different sensor data is maintained through radiation correction.

[0089] Based on the stratified sampling strategy, a network of ground plots covering different grassland types and degradation levels is constructed, and data including vegetation cover, species composition, biomass, and soil physical and chemical properties are collected to form a standardized grassland survey dataset.

[0090] The ground measured data and high-resolution remote sensing images are jointly analyzed to establish a correlation database of grassland types and degradation levels.

[0091] In summary, through fine data processing and the combination of multi-source remote sensing data, the accuracy and spatio-temporal adaptability of grassland monitoring are significantly improved. First, high-resolution optical satellite images are used for grassland mapping, combined with medium and low-resolution optical satellite images for long-term dynamic monitoring, and microwave remote sensing data are used to penetrate clouds and observe grassland biomass and water content, providing multi-dimensional data support for comprehensive monitoring of grassland degradation. All images are processed to remove strips, repair broken lines, and reduce noise to ensure data quality. Second, high-precision control point networks are used for geometric precision correction of remote sensing images to ensure pixel-level registration between different data sources, thereby improving the spatial consistency of different sensor data. In terms of image processing, atmospheric correction of optical remote sensing images eliminates atmospheric effects, and terrain correction of mountain grassland areas removes terrain shadows and slope effects, thereby improving the accuracy of ground information extraction. In addition, by combining high spatial resolution and high temporal resolution data, a data cube with unified time steps and multi-spatial scales is constructed, and the spectral consistency between different sensors is maintained through radiation correction, providing stable data support for long-term monitoring. Based on the stratified sampling strategy, a network of ground plots covering different grassland types and degradation levels is constructed, and data including vegetation cover, species composition, biomass, and soil physical and chemical properties are collected to form a standardized grassland survey dataset. Finally, through joint analysis of ground measured data and high-resolution remote sensing images, a correlation database of grassland types and degradation levels is established, providing high-precision and high-consistency basic data for grassland degradation monitoring and management.

[0092] Further, the atmospheric correction of optical remote sensing images uses an improved 6S or FLAASH atmospheric radiation transfer model, with parameters including atmospheric visibility 20-50 km, relative humidity 40-70%, aerosol optical thickness 0.05-0.5, and ozone content 250-350 Dobson units. The correction process combines near-surface meteorological station data to ensure a correction accuracy RMSE of less than 0.05 reflectance units.

[0093] Further, the ground plot network based on the stratified sampling strategy is designed as follows: 8-15 main plots are set for each grassland type, and 3-5 secondary plots are set in each main plot; the main plot is 30m x 30m in size, which matches the pixel size of medium-resolution remote sensing data; the secondary plot is 1m x 1m in size, and each secondary plot is arranged at an interval of 5-10m; the plot investigation indexes include vegetation coverage, dominant species composition, aboveground biomass, species richness, and soil surface structure.

[0094] Further, step 2 includes the following steps:

[0095] Based on the corrected multi-source remote sensing data, vegetation index, water index and soil index are calculated, and a comprehensive index set capable of comprehensively representing the structural and functional characteristics of grassland ecosystem is constructed;

[0096] For each index, the statistical characteristics in the time dimension and the spatial dimension are calculated, a multi-dimensional feature space is constructed, and the dynamic changes of the grassland ecosystem in different degradation stages are accurately described;

[0097] On the pixel, patch and landscape spatial scales, the information gain value and mutual information value of each index on grassland degradation classification are calculated, the classification contribution of each index at different scales is quantitatively evaluated, and a candidate index set with significant and stable contribution is selected through information stability analysis;

[0098] The response sensitivity of each index to the grassland degradation gradient is evaluated, the sensitivity coefficient and the discrimination index are calculated, and the performance differences of the index in different ecological regions, grassland types and degradation degree intervals are analyzed in combination with the true value library data, and the feature index with the most discrimination ability at each scale is identified;

[0099] The optimal index combination is automatically selected at different spatial scales to ensure the classification accuracy while minimizing the redundancy between indexes, and the economy and practicality of data acquisition are comprehensively considered to form an adaptive index combination scheme suitable for different scales and grassland types;

[0100] The index transfer relationship and conversion rules across scales are established, the classification performance and stability of the index system at each scale are evaluated based on independent validation data set, and finally a grassland monitoring index system with scale adaptability, regional pertinence and degradation sensitivity is constructed.

[0101] In summary, the accuracy and adaptability of grassland degradation monitoring are significantly improved through precise multi-source remote sensing data analysis and feature extraction. First, based on the corrected remote sensing data, vegetation index, water index and soil index are calculated, and a comprehensive index set is constructed to represent the structure and function of grassland ecosystem, providing multi-dimensional basic data for monitoring the ecological changes of grassland. Then, for each index, the statistical characteristics in time and space dimensions are calculated to construct a multi-dimensional feature space, accurately describing the dynamic changes of grassland ecosystem at different degradation stages. In terms of spatial scale analysis, the method calculates the information gain value and mutual information value of each index for grassland degradation classification, quantitatively evaluates the classification contribution of each index at different scales, and selects a candidate index set with significant contribution and stability through information stability analysis. In addition, the method also evaluates the response sensitivity of each index to the grassland degradation gradient, calculates the sensitivity coefficient and discrimination index, and analyzes the performance differences of the index in different ecological regions, grassland types and degradation degree intervals, to identify the most discriminant feature index. Finally, the method automatically selects the optimal index combination at different spatial scales, ensuring the classification accuracy and minimizing the redundant information, while considering the economy and practicality of data acquisition, forming an adaptive index combination scheme. By establishing the index transfer relationship and conversion rules across scales, and combining independent validation data set to evaluate the classification performance and stability, a grassland monitoring index system with scale adaptability, regional pertinence and degradation sensitivity is finally constructed.

[0102] Further, for each index, the statistical characteristics in time and space dimensions are calculated to construct a multi-dimensional feature space, accurately describing the dynamic changes of grassland ecosystem at different degradation stages, including the following steps:

[0103] For each index, the time mean μ t , time variance σ t 2 , change rate V t and autocorrelation R t are calculated to capture the stability, volatility and trend information of the grassland ecosystem at different times;

[0104] At each time, the spatial mean μ s , spatial variance σ s 2 and spatial autocorrelation I s are calculated for all index data to quantify the balance, dispersion and adjacency of the ecosystem in spatial distribution;

[0105] The statistical characteristics extracted in time and space dimensions are combined into a multi-dimensional feature vector F = [μ t , σ t2 V t ,R t ,μ s ,σ s 2 ,I s This forms a characteristic space that comprehensively depicts the dynamic changes in grassland degradation;

[0106] By comparing the feature vectors of different periods or regions and calculating the ecosystem dynamic change index D, the accurate determination and dynamic monitoring of the degradation stage of grassland ecosystems can be achieved. The formula is expressed as: Where, μ t (t) represents the mean of the index over time, indicating the mean of the grassland degradation index calculated at time point t; μ s (s) represents the mean of the index in the spatial dimension, specifically the mean of the grassland degradation index calculated in spatial unit s; μ t (t+1) represents the mean of the index over time, specifically the mean of the grassland degradation index calculated at time point t+1; μ s (s+1) represents the mean of the index in the spatial dimension, which is the mean of the grassland degradation index calculated in the spatial unit s+1.

[0107] Furthermore, at the three spatial scales of pixels, patches, and landscape, the information gain and mutual information values ​​of each indicator for grassland degradation grading are calculated to quantitatively assess the classification contribution of each indicator at different scales. Information stability analysis is then used to screen out a set of candidate indicators that contribute significantly and stably, including the following steps:

[0108] Based on remote sensing data of the target area, the spatial scales are divided into pixels, patches, and landscape.

[0109] At each spatial scale, collect the corresponding grassland degradation classification labels and remote sensing index data to ensure that the dataset at each scale contains sufficient sample size and spatial information.

[0110] For each spatial scale, the information gain formula IG(T,A)=H(T)-H(T∣A) is used to calculate the information gain value of each remote sensing index A for grassland degradation grade T. Here, IG(T,A) represents the information gain of remote sensing index A for grassland degradation grade T, that is, the degree to which the uncertainty of grassland degradation grade is reduced under the condition that A is known; H(T) represents the entropy of grassland degradation grade T, which measures the overall uncertainty of T; H(T∣A) represents the conditional entropy of grassland degradation grade T after the remote sensing index A is given, that is, the residual uncertainty of T under the condition that A is known.

[0111] The mutual information values between the remote sensing indexes and the grassland degradation classification at different spatial scales are calculated using the mutual information formula MI(A, T) = H(T) + H(A) - H(T, A), where MI(A, T) represents the mutual information of the remote sensing index A and the grassland degradation classification T, H(A) represents the entropy of the remote sensing index A, and H(T, A) represents the joint entropy of the grassland degradation classification T and the remote sensing index A.

[0112] The classification contribution degrees of the remote sensing indexes at different spatial scales are quantitatively evaluated through the information gain values and the mutual information values.

[0113] The standard deviations of the information gain values or the mutual information values of the remote sensing indexes at different spatial scales are calculated, and the formula is used. The stability of the information is evaluated, where μ IG represents the mean value of the information gain values calculated at different spatial scales, and is used to evaluate the classification contribution degree of the index as a whole; and σ IG represents the standard deviation of the information gain values calculated at different spatial scales, and is used to measure the stability of the classification contribution degree of the index at each scale.

[0114] According to the information stability analysis results, a set of remote sensing indexes that are stable and have significant contribution at different scales is screened out.

[0115] Further, step 3 includes the following steps:

[0116] The climate factor data and the human factor data are integrated to establish a database of potential driving forces of grassland degradation that is consistent in time and space, so as to ensure that the data of each driving factor is matched with the remote sensing monitoring data in terms of time and spatial resolution.

[0117] A multi-level Bayesian network model is constructed with the grassland degradation index as the response variable and the climate and human factors as the prediction variables, where the first level represents the response at the pixel scale, the second level represents the response at the patch scale, and the third level represents the response at the landscape scale, and each level is connected through conditional probability relationships.

[0118] The main effects, the interaction effects, and the contribution rates of the climate and the human factors at different scales are calculated, and the partial correlation analysis and the variance decomposition technique are used to separate the independent contributions of each factor.

[0119] The spatial scale conversion threshold of the dominant effects of the climate and the human factors is identified by comparing the information criteria and the Bayesian factors, and the optimal scale segmentation point is determined using the entropy maximization criterion to construct a scale-driving force response curve for identifying key scale nodes.

[0120] The relative contribution rates of each driving factor to the grassland degradation at different scales and the spatial differentiation characteristics thereof are calculated to generate a scale-driving force contribution rate map, so as to realize the precise quantification and spatial visualization of the influences of the climate and the human factors.

[0121] In summary, by integrating climate and human factors data and combining remote sensing monitoring data, a spatiotemporally consistent database of potential driving forces of grassland degradation is successfully constructed, thereby ensuring the matching of each driving factor data and remote sensing data in terms of time and spatial resolution. By constructing a multi-level Bayesian network model, the method accurately simulates the response of grassland degradation at different spatial scales (pixels, patches, and landscapes), with each level of the model being connected through conditional probability relationships. Furthermore, the method calculates the main effects, interaction effects, and contribution rates of climate and human factors at different scales, uses partial correlation analysis and variance decomposition techniques to isolate the independent contribution of each factor, and improves the accuracy of the analysis. By comparing the information criterion and the Bayesian factor, the spatial scale conversion threshold of the dominant effect of climate and human factors is identified, and the optimal scale segmentation point is determined using the entropy maximization criterion, a scale-driving force response curve is constructed, and key scale nodes are identified. In addition, the method calculates the relative contribution rate of each driving factor to grassland degradation at different scales and its spatial differentiation characteristics, generates a scale-driving force contribution rate map, and realizes the accurate quantification and spatial visualization of the influence of climate and human factors.

[0122] Further, the climate factors include average temperature, precipitation, evapotranspiration, and extreme climate events; the human factors include grazing pressure, farming and pastoral activities, infrastructure construction, and protection policies.

[0123] Further, the construction process of the multi-level Bayesian network model includes: in the structure learning stage, the network topology structure is determined by combining the greedy search algorithm and expert knowledge constraints; in the parameter learning stage, the expectation maximization algorithm is used to estimate the conditional probability table parameters; the evaluation indicators of the network model include AUC value, Kappa coefficient, and average log-likelihood value; the training-validation division ratio is 7:3, and the k-fold cross-validation is used to evaluate the stability of the model.

[0124] Further, the identification of the key scale node uses the following criteria: calculate the Bayesian factor B at each candidate scale, the formula is: B = P(D|M1) / P(D|M2), where M1 is the climate factor dominant model; M2 is the human factor dominant model; P(D|M1) represents the possibility of the occurrence of observation data D under the climate factor dominant model M1; P(D|M2) represents the possibility of the occurrence of observation data D under the human factor dominant model M2; when 2≤|logB|≤6, it is moderate evidence, and when |logB|>6, it is strong evidence, and the scale point with the maximum |logB| value is taken as the key conversion node; at the same time, the significance of the scale conversion point is verified by multi-scale variance analysis.

[0125] Further, the main effects, interaction effects and their contribution rates at different scales of climate and human factors are calculated, and the independent contribution of each factor is stripped off by using partial correlation analysis and variance decomposition technology, including the following steps:

[0126] Using a multiple linear regression model, setting the grassland degradation degree as the dependent variable, and the climate factors and human factors as the independent variables, the main effect coefficient is obtained by regression analysis, and the contribution rate of each main effect to the total variance is calculated;

[0127] The interaction term of climate factors and human factors is added to the regression model, the interaction effect coefficient is calculated, and the contribution rate of the interaction effect is calculated by variance analysis to measure the influence of the interaction of climate and human factors on grassland degradation;

[0128] The partial correlation coefficient and respectively assess the independent contribution of climate and human factors to grassland degradation under the control of another factor, wherein, represents the partial correlation coefficient between variable X1 and dependent variable Y after controlling variable X2; represents the partial correlation coefficient between variable X2 and dependent variable Y after controlling variable X1;

[0129] By applying the variance decomposition method, the total variance is decomposed into the variance of climate factors, human factors, interaction effects and error terms;

[0130] By calculating the variance contribution rate of each part, the contribution degree of each factor to the variation of grassland degradation is determined.

[0131] Further, step 4 includes the following steps:

[0132] At the pixel scale, local texture features are extracted using an improved convolutional neural network, at the patch scale, attention mechanism is integrated to enhance key area feature expression, and at the landscape scale, graph convolution network is introduced to capture the topological relationship between landscape units;

[0133] The spatial context information is used to optimize the classification boundary, and high-precision grassland degradation classification at 1-30 meter resolution is realized;

[0134] High-resolution remote sensing images are segmented at multiple scales to generate homogeneous patches with ecological significance, realize degradation classification at the function unit level of grassland, and analyze the influence of grassland fragmentation and connectivity changes on ecological function;

[0135] Capture the macro pattern and functional changes of grassland ecosystems, analyze the spatial aggregation and diffusion trend of degraded areas, and evaluate the integrity of the ecosystem;

[0136] Based on the multi-period classification results, the degradation transition matrix is constructed to quantitatively calculate the area proportion, spatial distribution characteristics and time evolution law of each degradation level at different scales.

[0137] In summary, by introducing advanced deep learning and graph neural network technology, the accuracy and spatio-temporal adaptability of grassland degradation level identification are significantly improved. At the pixel scale, an improved convolutional neural network is used to extract local texture features. At the patch scale, an attention mechanism is integrated to enhance the expression of key area features. At the landscape scale, a graph convolution network is introduced to capture the topological relationship between landscape units, thereby better representing the spatial structure of the grassland ecosystem. The use of spatial context information optimizes the classification boundary, enabling high-precision grassland degradation level identification at resolutions ranging from 1 to 30 meters. Meanwhile, by multi-scale segmentation of high-resolution remote sensing images, homogeneous patches with ecological significance are generated, further enabling degradation classification at the functional unit level of grassland, and analyzing the impact of grassland fragmentation and connectivity changes on ecological function. This method can capture the macro pattern and functional changes of the grassland ecosystem, analyze the spatial aggregation and diffusion trends of degraded areas, and assess the integrity of the ecosystem. In addition, based on the multi-period classification results, the degradation transition matrix is constructed to quantitatively calculate the area proportion, spatial distribution characteristics and time evolution law of each degradation level at different scales, providing comprehensive quantitative analysis support for grassland protection and restoration.

[0138] Further, the improved convolutional neural network is based on ResNet50 or DenseNet121 architecture, combined with depth separable convolution to reduce network parameter quantity by 40-60%; the network input includes 10-12 band remote sensing images and 3-5 terrain derivative indicators; the training process uses Adam optimizer with learning rate of 0.0005-0.002, batch size of 16-64, training rounds of 50-200, and cosine annealing learning rate strategy and early stopping method to avoid overfitting;

[0139] The attention mechanism uses channel and spatial dual attention modules, where channel attention extracts channel feature importance through the combination of global average pooling and max pooling, and spatial attention generates a spatial weight map through a 7x7 convolution; the two attention modules are applied in series to the feature map to improve the model's adaptive learning ability for different bands and spatial regions.

[0140] Further, step 5 includes the following steps:

[0141] The conversion rule library between the three scales of pixels, patches and landscapes is established, wherein: for discrete classification results, the dominant type method and weighted voting method are used for up-scale aggregation; for continuous degradation index, the regional average method and area weighted method are used for scale conversion; at the same time, a nonlinear conversion function considering spatial heterogeneity is designed to capture the mutation effect of degradation characteristics in the scale conversion process;

[0142] The global information of high-level scale is taken as prior knowledge to constrain the local detail expression of lower-level scale, and at the same time, the high-precision information of lower-level scale is taken as evidence to update the overall cognition of upper-level scale, so as to build a two-way information flow mechanism.

[0143] The entropy, variance and confidence interval of each spatial unit at different scales are calculated to identify the high uncertainty area and analyze the main factors causing the uncertainty.

[0144] The grassland types and degradation processes highly sensitive to specific scales are identified, the scale adaptability index is calculated, the applicability of monitoring results at each scale is quantitatively evaluated, and the optimal scale selection decision tree is constructed.

[0145] The monitoring results at each scale are fused according to the reliability weight to generate the comprehensive monitoring results of grassland degradation with unified precision standard and complete spatial coverage.

[0146] In summary, by establishing the multi-scale conversion rule library, the accurate conversion between the scales of pixels, patches and landscapes is realized, and the spatial adaptability and precision of grassland degradation monitoring are significantly improved. For discrete classification results, the dominant type method and weighted voting method are used for scale up aggregation; for continuous degradation index, the regional average method and area weighted method are used for scale conversion. At the same time, a nonlinear conversion function considering spatial heterogeneity is designed, which can capture the mutation effect of degradation characteristics in the scale conversion process. This method also builds a two-way information flow mechanism, which uses the global information of high-level scale to constrain the local detail expression of lower-level scale, and at the same time, uses the high-precision information of lower-level scale to update the overall cognition of upper-level scale, ensuring the effective fusion of information at each scale. By calculating the entropy, variance and confidence interval of each spatial unit at different scales, the high uncertainty area is identified, and the main factors causing the uncertainty are analyzed, further improving the reliability of the monitoring results. This technology can identify the grassland types and degradation processes highly sensitive to specific scales, calculate the scale adaptability index, quantitatively evaluate the applicability of monitoring results at each scale, and finally construct the optimal scale selection decision tree. By fusing the monitoring results at each scale and weighting according to the reliability weight, the comprehensive monitoring results of grassland degradation with unified precision standard and complete spatial coverage are generated.

[0147] Further, the specific implementation of the constructed bidirectional information flow mechanism includes: the top-down constraint flow adopts a hierarchical conditional random field model for spatial context constraints; the bottom-up update flow adopts an aggregation-decomposition strategy to realize information uploading through majority voting, regional statistics, and boundary preservation filtering; the information flow in both directions is iterated alternately 2-3 times until the classification result change rate of adjacent iterations is less than 5% or the maximum iteration number is reached 5 times; the consistency index Kappa between scales is required to reach above 0.65.

[0148] Further, step 6 includes the following steps:

[0149] Based on the grassland degradation level classification results and spatial statistical analysis, a high-precision degradation level distribution map is generated, and the spatial distribution characteristics of different degradation degree regions are intuitively displayed through a graded color scale;

[0150] A spatial contribution rate model of climate factors and human factors is constructed to generate a driving force analysis map, and a double-factor color coding method is used to intuitively present the spatial heterogeneity of the dominant factors;

[0151] Combined with historical data sequences and current monitoring results, a grassland degradation risk warning map is generated to identify potential rapid degradation areas and ecological fragile points, and according to the risk level, corresponding warning thresholds are set;

[0152] According to the grassland degradation degree, driving factors, and ecological sensitivity, management units are divided, and a zoning management suggestion map is generated to propose targeted management measures to achieve the efficiency and pertinence of grassland protection and restoration.

[0153] In summary, through the grassland degradation level classification results and spatial statistical analysis, a high-precision degradation level distribution map is generated to intuitively display the spatial distribution characteristics of different degradation degree regions. Combined with the spatial contribution rate model of climate factors and human factors, a driving force analysis map is generated, and through a double-factor color coding method, the spatial heterogeneity of the dominant factors is clearly presented. By combining historical data and current monitoring results, a grassland degradation risk warning map is generated to identify potential rapid degradation areas and ecological fragile points, and to provide warning thresholds for risk management. In addition, based on the grassland degradation degree, driving factors, and ecological sensitivity, management units are divided and a zoning management suggestion map is generated to propose targeted management measures to achieve the efficiency and pertinence of grassland protection and restoration.

[0154] As shown in Figure 2 The grassland degradation degree monitoring model based on remote sensing technology includes the following modules:

[0155] The data acquisition and preprocessing module is responsible for integrating multi-source remote sensing data and ground measured data, constructing a spatio-temporally consistent multi-scale dataset, and performing preprocessing to ensure data accuracy and consistency.

[0156] The degradation degree evaluation module calculates relevant indicators and performs multi-dimensional feature analysis based on the preprocessed remote sensing data, and selects the optimal index combination.

[0157] The driving force analysis module quantifies the main effects, interaction effects and contribution rates of climate and human factors on grassland degradation, and identifies key scale nodes.

[0158] The result display and management module displays the grassland degradation grade distribution, driving force analysis, risk warning and management suggestions in the form of intuitive charts, and performs management.

[0159] The multi-scale analysis and conversion module establishes conversion rules between different spatial scales, realizes multi-scale segmentation and feature extraction, constructs a bidirectional information flow mechanism, and ensures reliable conversion and fusion of analysis results at different scales.

[0160] The management strategy generation module automatically generates differentiated management strategies based on grassland degradation monitoring results and related analysis, proposes targeted management measures, and realizes the efficiency and pertinence of grassland protection and restoration.

[0161] In summary, the grassland degradation monitoring model based on remote sensing technology realizes accurate monitoring and management of grassland degradation through the collaborative work of multiple modules. The data acquisition and preprocessing module integrates multi-source remote sensing and ground data, constructs a spatio-temporally consistent multi-scale dataset, and ensures data accuracy and consistency. The degradation degree evaluation module selects the optimal remote sensing index through multi-dimensional feature analysis to accurately evaluate the degree of grassland degradation. The driving force analysis module quantifies the main effects and interaction effects of climate and human factors on degradation, and identifies key scale nodes. The result display and management module presents the grassland degradation grade, driving force analysis, risk warning and management suggestions through intuitive charts, providing a basis for decision-making. The multi-scale analysis and conversion module realizes the conversion and fusion of analysis results at different spatial scales, and constructs a bidirectional information flow mechanism to ensure effective transmission of information between scales. The management strategy generation module automatically generates differentiated management strategies based on monitoring results, and proposes efficient and targeted measures for grassland protection and restoration.

[0162] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the same. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring the degree of grassland degradation based on remote sensing technology, characterized in that, The method comprises the following steps: Step 1, constructing a spatio-temporally consistent multi-scale remote sensing dataset, and combining ground measured data to establish a high-precision true value library of grassland types and degradation degrees; Step 2, automatically screening remote sensing indexes most sensitive to grassland degradation, and constructing a remote sensing index system suitable for different spatial scales; Step 3, identifying key scale conversion nodes driven by climate and human factors for grassland degradation; Step 4, realizing grassland degradation grading from pixel scale to landscape scale, and quantitatively analyzing the spatial distribution, area change and spatio-temporal evolution trend of each degradation grade; Step 5, realizing reliable conversion and up-down transmission of analysis results of different scales, and generating comprehensive grassland degradation monitoring results; Step 6, automatically generating differentiated management strategies, and outputting grassland degradation grade distribution map, driving force analysis map, risk warning map and partition management suggestion map; Step 2 comprises the following steps: Based on the corrected multi-source remote sensing data, the vegetation index, water index and soil index are calculated, and a comprehensive index set capable of fully representing the structural and functional characteristics of the grassland ecosystem is constructed; For each index, the statistical characteristics in the time dimension and the spatial dimension are calculated, a multi-dimensional feature space is constructed, and the dynamic changes of the grassland ecosystem at different degradation stages are accurately described; On the pixel, patch and landscape three spatial scales, the information gain value and mutual information value of each index to the grassland degradation grading are calculated, the classification contribution degree of each index at different scales is quantitatively evaluated, and the candidate index set with significant contribution and stability is screened out through information stability analysis; The response sensitivity of each index to the grassland degradation gradient is evaluated, the sensitivity coefficient and the discrimination index are calculated, and the performance difference of the index in different ecological regions, grassland types and degradation degree intervals is analyzed in combination with the true value library data, and the feature index with the most distinguishing ability at each scale is identified; The optimal index combination is automatically selected on different spatial scales to ensure the classification accuracy while minimizing the redundancy between indexes, and the economic and practicality of data acquisition are comprehensively considered to form an adaptive index combination scheme suitable for different scales and grassland types; The cross-scale index transmission relationship and conversion rule are established, the classification performance and stability of the index system at each scale are evaluated based on the independent verification data set, and finally a grassland monitoring index system with scale adaptability, regional pertinence and degradation sensitivity is constructed; Step 3 comprises the following steps: Integrating climate factors and human factors data, a spatio-temporally consistent grassland degradation potential driving force database is established to ensure that the data of each driving factor is matched in time and spatial resolution with the remote sensing monitoring data; Taking the grassland degradation index as the response variable and the climate and human factors as the prediction variables, a multi-level Bayesian network model is constructed, wherein: the first level represents the pixel scale response, the second level represents the patch scale response, and the third level represents the landscape scale response; each level is connected through conditional probability relationship; The main effect, interaction effect and contribution rate of climate and human factors at different scales are calculated, and the independent contribution of each factor is stripped by using partial correlation analysis and variance decomposition technology; In combination with information criterion and Bayesian factor comparison, the spatial scale transition threshold of dominant effect of climate and human factors is identified, and the optimal scale segmentation point is determined by using entropy maximization criterion to construct scale-driving force response curve for identifying key scale nodes. The relative contribution rate of each driving factor to grassland degradation at different scales and its spatial differentiation characteristics are calculated to generate scale-driving force contribution rate atlas, realizing the accurate quantification and spatial visualization of the influence of climate and human factors.

2. The remote sensing technology-based grassland degradation degree monitoring method according to claim 1, characterized in that, For each index, the statistical characteristics in the time and space dimensions are calculated to construct a multi-dimensional feature space, accurately depicting the dynamic changes of grassland ecosystem at different degradation stages, including the following steps: For each index, the time mean μ t , the time variance σ t 2 , the rate of change V t and the autocorrelation R t are calculated to capture the stability, volatility and trend information of the grassland ecosystem at different times. At each time, the spatial mean μ is calculated for all index data s , the spatial variance σ s 2 and the spatial autocorrelation I s to quantify the balance, dispersion and adjacency of the ecosystem in the spatial distribution; The statistical features extracted from the time and space dimensions are combined into a multi-dimensional feature vector F = [μ t ,σ t 2 ,V t ,R t ,μ s ,σ s 2 ,I s ], forming a feature space that comprehensively characterizes the dynamic changes of grassland degradation; The dynamic index D of the ecosystem is calculated by comparing the eigenvectors of different periods or regions, and the degradation stage of the grassland ecosystem is accurately determined and dynamically monitored, and the formula is expressed as: Wherein, μ t (t) is the mean value of the index in the time dimension, which represents the mean value of the grassland degradation index calculated at time point t; μ s (s) is the mean value of the index in the space dimension, which represents the mean value of the grassland degradation index calculated at space unit s; μ t (t+1) is the mean value of the index in the time dimension, which represents the mean value of the grassland degradation index calculated at time point t+1; μ s (s+1) is the mean value of the index in the space dimension, which represents the mean value of the grassland degradation index calculated at space unit s+1. 3.The remote sensing-based grassland degradation degree monitoring method according to claim 1, characterized in that, At the pixel, patch and landscape scales, the information gain value and mutual information value of each index to grassland degradation classification are calculated to quantitatively evaluate the classification contribution of each index at different scales, and the candidate index set with significant and stable contribution is selected through information stability analysis, including the following steps: According to the remote sensing data of the target area, the pixel, patch and landscape scales are divided; At each spatial scale, the corresponding grassland degradation classification labels and remote sensing index data are collected to ensure that the data set at each scale contains sufficient sample size and spatial information; For each spatial scale, the information gain formula IG(T,A)=H(T)-H(T|A) is used to calculate the information gain value of each remote sensing index A to grassland degradation classification T, where IG(T,A) represents the information gain of remote sensing index A to grassland degradation classification T, i.e. the degree of reduction in the uncertainty of grassland degradation classification under the condition of knowing A; H(T) represents the entropy of grassland degradation classification T, measuring the overall uncertainty of T; H(T|A) represents the conditional entropy of grassland degradation classification T given remote sensing index A, i.e. the remaining uncertainty of T under the condition of knowing A; The mutual information formula MI(A,T)=H(T)+H(A)-H(T,A) is used to calculate the mutual information value between remote sensing index and grassland degradation classification at different spatial scales, where MI(A,T) represents the mutual information between remote sensing index A and grassland degradation classification T; H(A) represents the entropy of remote sensing index A; H(T,A) represents the joint entropy of grassland degradation classification T and remote sensing index A; The classification contribution of each remote sensing index at different spatial scales is quantitatively evaluated through information gain value and mutual information value; Calculate the standard deviation of information gain or mutual information values ​​of remote sensing indicators at different spatial scales, and use the formula: To assess the stability of the information, where μ IG σ represents the mean of the information gain values ​​calculated at different spatial scales, used to evaluate the overall classification contribution of the indicator; IG The standard deviation of the information gain value calculated at different spatial scales is used to measure the stability of the classification contribution of the index at each scale. According to the information stability analysis results, the remote sensing index set with stable performance and significant contribution at different scales is selected. 4.The remote sensing-based grassland degradation degree monitoring method according to claim 1, characterized in that, The main effect, interaction effect and contribution rate of climate and human factors at different scales are calculated, and the independent contribution of each factor is stripped using partial correlation analysis and variance decomposition techniques, including the following steps: Using a multiple linear regression model, set the grassland degradation degree as the dependent variable, and the climate factors and human factors as the independent variables, obtain the main effect coefficient through regression analysis and calculate the contribution rate of each main effect to the total variance; The interaction term of climate factors and human factors is added to the regression model to calculate the interaction effect coefficient, and the contribution rate of the interaction effect is calculated through variance analysis to measure the influence of the interaction of climate and human factors on grassland degradation. By calculating the partial correlation coefficient and respectively assess the independent contribution of climate and human factors to grassland degradation, where denotes the partial correlation coefficient between the variable after controlling for the variable and the dependent variable ; denotes the partial correlation coefficient between the variable after controlling for the variable and the dependent variable ; The total variance was decomposed into the variance of climate factors, human factors, interaction effects, and error terms by variance decomposition method. The contribution of each factor to the variation of grassland degradation was determined by calculating the variance contribution rate of each part. 5.The remote sensing-based grassland degradation degree monitoring method according to claim 1, characterized in that, Step 4 includes the following steps: Local texture features were extracted at the pixel scale using an improved convolutional neural network, key region feature expression was enhanced at the patch scale by integrating attention mechanism, and graph convolution network was introduced at the landscape scale to capture the topological relationship between landscape units. The spatial context information was used to optimize the classification boundary, and high-precision grassland degradation classification was achieved at 1-30 meter resolution. High-resolution remote sensing images were segmented at multiple scales to generate homogeneous patches with ecological significance, and degradation classification was realized at the functional unit level of grassland, and the impact of grassland fragmentation and connectivity changes on ecological function was analyzed. The macro pattern and functional changes of grassland ecosystem were captured, the spatial aggregation and diffusion trend of degraded areas were analyzed, and the integrity of the ecosystem was evaluated. Based on the multi-period classification results, a degradation transition matrix was constructed, and the area proportion, spatial distribution characteristics, and temporal evolution law of each degradation level at different scales were quantitatively calculated. 6.The remote sensing-based grassland degradation degree monitoring method according to claim 1, characterized in that, Step 5 includes the following steps: A conversion rule library was established between the pixel, patch, and landscape scales. For discrete classification results, the dominant type method and weighted voting method were used for upscaling aggregation. For continuous degradation index, the regional average method and area weighting method were used for scale conversion. At the same time, a nonlinear conversion function considering spatial heterogeneity was designed to capture the mutation effect of degradation characteristics in the scale conversion process. The global information at high-level scale was used as prior knowledge to constrain the local detail expression at lower-level scale, and the high-precision information at lower-level scale was used as evidence to update the overall cognition at higher-level scale, and a bidirectional information flow mechanism was constructed. The entropy, variance, and confidence interval of each spatial unit at different scales were calculated to identify high-uncertainty areas and analyze the main factors causing uncertainty. The grassland types and degradation processes that are highly sensitive to a particular scale were identified, the scale adaptability index was calculated, the applicability of monitoring results at each scale was quantitatively evaluated, and an optimal scale selection decision tree was constructed. The monitoring results at each scale were fused according to the reliability weight to generate a unified precision standard and complete spatial coverage of grassland degradation monitoring comprehensive results. 7.The remote sensing-based grassland degradation degree monitoring method according to claim 1, characterized in that, Step 6 includes the following steps: Based on the grassland degradation classification results and spatial statistical analysis, a high-precision degradation level distribution map was generated, and the spatial distribution characteristics of different degradation levels were visually displayed through hierarchical color coding. A spatial contribution rate model of climate factors and human factors was constructed to generate a driving force analysis map, and a two-factor color coding method was used to visually present the spatial heterogeneity of dominant factors. Combined with historical data series and current monitoring results, a grassland degradation risk warning map was generated to identify potential rapid degradation areas and ecological fragile points, and corresponding warning thresholds were set according to the risk level. According to the degree of grassland degradation, driving factors, and ecological sensitivity, management units were divided, and a zoning management suggestion map was generated to propose targeted management measures for efficient and targeted grassland protection and restoration.

8. A model for monitoring the degree of grassland degradation based on remote sensing technology, used to implement the method for monitoring the degree of grassland degradation based on remote sensing technology according to any one of claims 1-7, characterized in that, The following modules are included: Data acquisition and preprocessing module, responsible for integrating multi-source remote sensing data and ground measured data, constructing spatio-temporal consistent multi-scale data set, and preprocessing to ensure data accuracy and consistency; Degradation assessment module, based on the preprocessed remote sensing data, calculates relevant indicators and conducts multi-dimensional feature analysis, selects the optimal index combination; Driving force analysis module, quantifies the main effect, interaction effect and contribution rate of climate and human factors on grassland degradation, and identifies key scale nodes; Result display and management module, displays the grassland degradation grade distribution, driving force analysis, risk warning and management suggestions in the form of intuitive charts, and manages them; Multi-scale analysis and conversion module, establishes conversion rules between different spatial scales, realizes multi-scale segmentation and feature extraction, constructs a two-way information flow mechanism, and ensures reliable conversion and fusion of analysis results of different scales; Management strategy generation module, generates differentiated management strategies and proposes targeted management measures based on grassland degradation monitoring results and related analysis, to achieve efficiency and pertinence of grassland protection and restoration.

Images