A method for identifying and evaluating key nodes of an ecological system in a northern arid region

Abstract

A kind of northern arid zone ecosystem key node identification and evaluation method, steps include: calling long time series remote sensing data, extract ecological index from the called remote sensing data to construct ecological risk index;Based on ecological space network analysis method, using degree centrality DC as node importance evaluation index, the connection ability of network node is quantitatively evaluated, and the node whose DC is located in the top 10% is screened as the main key ecological node;According to the direct flow matrix and the comprehensive flow matrix between nodes, the risk flow intensity is graded, the main control path, the risk diffusion path and the key connection path in the network are extracted, and the key ecological path is obtained;Based on ecological risk, key node and key path, an ecological comprehensive evaluation system is constructed, the ecological partition is identified, and the dynamic closed loop evaluation of ecological system is realized.The present application can realize the rapid and accurate identification of key nodes of arid zone ecosystem, and obtain the key ecological path, realize the identification and dynamic closed loop update of repair area.

Description

Technical Field

[0001] This invention relates to a method for identifying and evaluating key nodes in an ecosystem in arid northern regions, belonging to the field of ecological environment monitoring and ecological network analysis technology. Background Technology

[0002] Ecological nodes (also known as stepping stones) are points in an ecological network that connect adjacent ecological patches and play a crucial role in the flow of ecological energy. On some landscape surfaces, ecological nodes are intersections or turning points of corridors. The construction of ecological nodes will effectively improve the overall connectivity of the regional landscape and promote the healthy cycle of ecological functions. How to promptly identify the most critical and vulnerable ecological nodes in ecosystems under large-scale, complex environments is an urgent problem to be solved in ecological protection and governance.

[0003] Traditional ecological assessments often focus on grid-scale mean statistics or single-indicator monitoring, failing to reflect the network structure characteristics and risk propagation pathways of ecosystems. Furthermore, existing research methods frequently focus only on individual ecological indicators or risk factors, lacking a systematic analysis of the complete chain of "risk source—risk receptor—exposure response," making it difficult to accurately reflect the driving mechanisms of ecological degradation. On the other hand, ecosystems exhibit distinct spatial network attributes, with different patches forming ecological connections through energy flow, material exchange, and biological migration. However, existing methods lack a unified approach for quantifying node importance. Traditional node analysis often relies solely on structural centrality indicators (such as degree centrality and betweenness), neglecting attributes like ecosystem service functions and environmental risk exposure levels, resulting in identification results that fail to comprehensively reflect the integrated role of nodes within the ecosystem. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying and evaluating key nodes in ecosystems in arid northern regions. This method can achieve rapid and accurate identification of key nodes in ecosystems in arid regions, obtain key ecological pathways, and realize the identification of restoration areas and dynamic closed-loop updates.

[0005] To achieve the above objectives, this invention provides a method for identifying and evaluating key nodes in ecosystems in arid northern regions, comprising the following steps: S1. Ecological Risk Identification: Based on Google Earth Engine (GEE), long-term remote sensing data is called, and ecological indicators are extracted from the called remote sensing data to construct an ecological risk index, so as to realize the quantitative inversion of the spatial pattern of ecological risk in large-scale arid areas. S2. Key Ecological Node Identification: Based on the ecological spatial network analysis method, the degree centrality DC is used as the node importance evaluation index to quantitatively evaluate the connectivity of network nodes, and the nodes with DC in the top 10% are selected as the main key ecological nodes. S3. Critical Path Identification: Based on the direct flow matrix and comprehensive flow matrix between nodes, the intensity of risk flows is classified, and the main control path, risk diffusion path and key connection path in the network are extracted to obtain the critical ecological path. S4. Comprehensive Evaluation: Based on ecological risks, key nodes and key paths, construct an ecological comprehensive evaluation system, identify ecological zones, and realize dynamic closed-loop evaluation of the ecosystem.

[0006] Furthermore, the specific process of S1 is as follows: S1.1 Data Selection: A spatially consistent, temporally coherent, and quality-controllable basic remote sensing dataset is constructed to provide data support for ERI construction and key ecological node identification. Specifically, medium- and high-resolution multi-source satellite remote sensing images are acquired online through a remote sensing data cloud platform as the main data foundation. A three-month time observation window in a typical season is used to obtain stable vegetation, habitat, and thermal information. To obtain stable statistical samples for model training, areas with vegetation cover, bare surfaces, water bodies, and impermeable surfaces with cloud cover ≤10% are selected within the study area as training areas. The training area is constructed with the geometric center of the study area as the base point, and a buffer zone of tens of kilometers is constructed outward. The final training area is formed after intersecting with the boundary of the study area. This training area is used to calculate subsequent quantile normalization parameters, image matching parameters, and statistical features such as the PCA covariance matrix, thereby improving the robustness of the model and reducing the impact of missing data in marginal areas. S1.2 Remote sensing image processing: S1.2-1. Using the quality control band, the bits corresponding to clouds, cloud shadows, and snow cover are analyzed to construct a logical mask of "non-cloud—non-cloud shadow—non-snow" to automatically remove contaminated pixels. Subsequently, based on a quantile linear matching strategy using the seasonal median benchmark, the true surface reflectance information is obtained, and the remote sensing image is radiometrically calibrated to eliminate the influence of sensor response differences. The formula is: ;in, Seasonal reference image in band The pixel value, For the first Scene image in band The pixel value, This represents the number of images within the same seasonal time window. It is a median composition operator; Using the median composite results of multiple images of the same season in the study area as a radiometric reference, the difference between the 2% and 98% quantiles of the image to be corrected and the reference image in the same band is calculated. A linear transformation function is constructed and applied to the whole image; thereby adaptively reducing the systematic radiometric bias caused by seasonal atmospheric transparency fluctuations, changes in observation angle and wind and sand environment in arid areas.

[0007] S1.2-2, A stripe suppression algorithm based on vertical direction decomposition extracts low-frequency background by applying a local smoothing kernel along the stripe direction. The original image is then decomposed into low-frequency and high-frequency components and reconstructed according to a preset ratio. This effectively reduces stripe and ripple noise while preserving surface texture and edge information, significantly improving the spatial consistency of multiple images. The calculation formula is as follows:

[0008] in, For low-frequency background components, This is the corrected image before band suppression. A one-dimensional smooth kernel along the strip direction. This represents the convolution / moving average operation;

[0009] in, To smooth the window length;

[0010] in, These are high-frequency components;

[0011] in, The output image after stripe suppression. This is the band suppression coefficient. The larger the value, the stronger the inhibition. S1.2-3. Perform seasonal median composite analysis on the images that have undergone the above processing within a preset time window. The calculation formula is as follows: ; in, For seasonal composite images in band The pixel value, This is the image after band suppression for the m-th scene. This represents the number of valid images within the time window. It is a median composition operator; ;in, For pixels Number of valid observations within the window The minimum effective observation threshold; By eliminating short-term weather disturbances and transient sensor noise, standardized synthetic images that can stably characterize the seasonal ecological state of the study area are obtained. These images are then used as a unified input for risk index calculation, principal component analysis weight extraction, and construction of the ecological risk index (ERI), thus achieving automated and homogenized conversion from raw multi-temporal images to a set of ecological risk factors.

[0012] S1.3 Risk Indicator Calculation: Risk indicators include risk receptor indicators, risk source indicators, and exposure response indicators. The calculation of each indicator is as follows: S1.3-1. Risk receptor indicators are used to characterize the health status of an ecosystem and its sensitivity to external stresses. Two types of indicators are selected: NDVI and WET. NDVI, based on red and near-infrared reflectance, accurately reflects vegetation cover and photosynthetic activity, and is the most sensitive indicator of vegetation degradation and recovery in arid regions. Median composites of seasonal images can stably present the overall vegetation status and avoid short-term weather disturbances. The calculation formula is as follows: ; in, For pixels Normalized differential vegetation index, Near-infrared reflectance, Red light reflectance, It is a very small constant; ; in, For seasonal NDVI, For the first NDVI, Median synthesis; WET, constructed using Tasseled Cap transformation, comprehensively reflects the overall moisture content of soil and vegetation. It can sensitively capture water deficit and soil drying processes in arid regions, possessing core monitoring value in arid ecosystems where water is the primary limiting factor. The calculation formula is as follows: ; in, For TasseledCap humidity component, Let reflectivity vector be the vector. This represents the humidity component coefficient vector corresponding to the sensor. For the first Each band coefficient Reflectivity of each band;

[0013] in, WET on a seasonal scale For the first JingWET, Median synthesis; S1.3-2. Risk source indicators are used to characterize the main negative driving forces leading to ecological degradation. The SI and BSI indicators are used together to reflect the core driving factors of land quality deterioration in arid areas. The SI utilizes the spectral differences between visible light and short-wave infrared to enhance salinity information, effectively identifying salinization areas caused by salt accumulation, salinization return, or improper irrigation. The calculation formula is as follows: ; in, The salt enhancement index, For shortwave infrared reflectivity, Near-infrared reflectance, It is a very small constant; The Baseline Indicator (BSI) is constructed using blue light, red light, near-infrared light, and short-wave infrared light to reflect the degree of bare land exposure. It is an important indicator for identifying the dynamics of soil desertification, wind erosion diffusion, and degradation edges. The calculation formula is as follows: ; in, The bare soil index, For shortwave infrared reflectivity, Red light reflectance, Near-infrared reflectance, Blue light reflectance, It is a very small constant; S1.3-3. Exposure response indicators characterize the immediate response of ecosystems to environmental stresses such as water deficit and thermal stress. The LST and ET indicators are selected. LST is retrieved using thermal infrared band inversion, revealing the thermal risks caused by increased bare surface area, decreased vegetation cooling capacity, and limited surface evaporation. The calculation formula is as follows: ; in, Surface temperature, unit: Brightness temperature, unit: The center wavelength of thermal infrared, For constant terms, Let be Planck's constant. At the speed of light, Boltzmann's constant; For pixels The surface emissivity, The conversion constant; The ET index is constructed using the inverse relationship between NDVI and LST to reflect vegetation water use capacity and ecosystem energy balance. The calculation formula is as follows: ; in, As an evapotranspiration response index, For the normalized positive NDVI term, The normalized LST is used; when ET decreases, it indicates a decline in system evapotranspiration and weakening of ecological functions, which is an important response signal to drought stress. S1.4 Risk Factor Normalization: Due to the significant differences in the dimensions and numerical ranges of different indicators, a quantile-robust normalization method is adopted to ensure the comparability of each indicator in subsequent PCA and comprehensive risk assessment, and to reduce the impact of extreme values ​​and local anomalies. Details are as follows: S1.4-1. To ensure consistency in the direction of indicators, the directions of indicators are unified according to their ecological significance: the larger the values ​​of NDVI, WET, and ET, the better the ecology, so they are taken as negative values, so the larger the values, the higher the risk; the larger the values ​​of SI, BSI, and LST, the higher the risk, so they are kept in their original directions. S1.4-2. Calculate the 2nd and 98th quantiles for each indicator within the training region to construct a robust interval; based on this interval, normalize the indicator values ​​to the following method. : ; in, As an indicator The normalization result, Indicators The training region is divided into 2% and 98% quantiles, and the data is limited to the effective range to prevent normalization distortion due to local anomalies or noise. When the quantile difference is too small or data is missing, minimum / maximum values ​​are used for backtracking normalization to ensure algorithm stability. The calculation formula is as follows: ; in, To revert to the normalization result, The index value after unifying the direction. These are the minimum and maximum values, respectively. It is a very small constant; after normalization, all six categories of indicators follow similar numerical ranges and have the same direction, providing a unified input for subsequent PCA weighting and ERI construction; S1.5, PCA Weight Extraction and Eigenvector Direction Constraint: The PCA method is used to extract the comprehensive risk gradient of six types of indicators, and a weight system for the ecological risk index is constructed, specifically as follows: S1.5-1. Construct the normalized indicators as an array, using pixels within the training region as samples. Obtain the correlation structure between indicators by calculating the sample covariance matrix. The calculation formula is as follows: ; in, The sample covariance matrix is ​​used to characterize the correlation structure among indicators. The centered sample matrix, This represents the number of training samples; To improve statistical robustness, a two-level strategy is adopted to estimate the sample covariance: when the number of training samples is sufficient and the covariance can be calculated, the sample covariance matrix is ​​used; when the number of samples is insufficient or the structure is abnormal, the calculation regresses to the region covariance matrix to ensure matrix invertibility and the safety of eigenvalue decomposition. The formula for the regression mechanism is as follows: ; in, This is the safety matrix used for eigenvalue decomposition. The region covariance matrix, The diagonal regularization coefficient is... It is the identity matrix; S1.5-2. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and eigenvectors of each principal component; select PC1 as the comprehensive risk gradient, whose eigenvector elements reflect the contribution of each indicator in the risk direction. The formulas for eigenvalue decomposition and PC1 are as follows: ; in, For the first 1 eigenvalue, For the corresponding feature vector, Principal component number, It is the covariance matrix; ; in, The variance explained by the first principal component, PC1. The largest eigenvalue, It is the sum of all eigenvalues; ; in, For pixels PC1 score, The eigenvector of PC1 is the first One element, For the first A normalized index value; To ensure consistency in ecological significance, a directional constraint is applied to the entire feature vector based on the weight sign of the NDVI_adj vegetation degradation index: when When the weights of the eigenvectors in the first principal component are negative, the overall eigenvector is multiplied by -1, making... The contribution is positive, ensuring that the larger the PC1 value, the higher the ecological risk. This yields the PCA weights corresponding to the six indicators, realizing a data-driven and consistent ecological risk weight system. S1.6 Construction of ERI: After obtaining the PCA weighting system, the six normalized risk indicators are weighted and superimposed according to their weights to obtain a preliminary ecological risk index: ; in, PCA weights To normalize the index value, and to enhance numerical stability, the ERI undergoes two-stage normalization: training area normalization, which standardizes the value using quantiles or extreme value ranges within the training area, ensuring it roughly conforms to the normal distribution within that area. The range is calculated using the following formula: ; in, The normalized risk index for the training area. Training areas The 2% and 98th percentiles, It is a very small constant; the study area is normalized as a whole: then it is normalized again according to the statistical interval of the entire study area, so that the final ERI is uniformly distributed within the study area. The interval is calculated using the following formula: ; in, This represents the final ecological risk index after normalization for the entire study area. The study area The 2% and 98th percentiles, It is a very small constant; the higher the ERI value, the greater the risk to the ecosystem, and it is used to reflect the comprehensive ecological risk level of multiple factors such as vegetation degradation, soil salinization, bare land expansion, and high temperature stress. S1.7 Construction of the Ecological Risk Spatial Network: After obtaining the spatial distribution of ERI, it is configured to divide the study area into spatial nodes according to a fixed grid, and each node is assigned the attributes of area, ecological type, ecosystem service level, and mean ERI. Edges are constructed based on the spatial relationships between nodes, generated through the minimum cumulative resistance (MCR) path based on the ecological resistance surface. Ecological resistance is provided by risk source indicators BSI, SI, and LST; the higher the risk, the greater the resistance. Risk receptors and exposure response indicators can reduce resistance. The formula for constructing the resistance surface is: ; in, This represents the pixel resistance value. It is the minimum ecological resistance constant. As a normalized indicator for risk source categories, For the normalized index of the sustained-release term, As the risk source weight, To mitigate the weight, Used to ensure that the resistance is not less than .

[0014] Step S1 can be used to construct a spatial connectivity network that conforms to the characteristics of ecological processes in arid regions, in order to identify nodes that play a key regulatory role in the ecological structure.

[0015] Furthermore, the specific process of S2 is as follows: S2.1 Calculate the node centrality index based on the ecological space network, and use the natural breakpoint method to... To classify and form Level of importance nodes; Level 1 nodes are core control nodes of the ecosystem, with high centrality, high service function level and high ERI value, and are key to maintaining regional ecological stability. Level nodes are important support nodes, and are mostly located in corridor intersection areas; Level nodes are general ecological nodes; Level 1 nodes are mostly located in the structural edge area or high-risk degradation nodes, and therefore have priority for repair. S2.2 Based on the node level, the study area is divided into the ecological protection priority area (i.e., Level I nodes and their affected areas), the corridor control area (i.e., Level II nodes and key corridor zones), the restoration priority area (i.e., Level IV nodes and their affected areas), and the general management area (i.e., Level III nodes and the remaining areas); forming a functionally clear and hierarchically distinct ecological spatial pattern, providing a basis for regional ecological governance and restoration.

[0016] Furthermore, the specific process of S3 is as follows: S3.1 Network Edge Construction: Based on ERI, an ecological resistance surface is constructed, and ecological corridors are identified through MCR paths to achieve a realistic simulation of potential connectivity structures. The uneven resistance in arid regions and the diffusion of ecological processes along corridors are fully considered. The formula for calculating the ecological resistance surface is as follows: ; in, For grid cells Ecological resistance value at the location; For grid cells The ecological risk index after quantile normalization; This is the minimum ecological resistance constant, used to characterize the case where ecological passage costs are minimized; This is the maximum ecological resistance constant, used to characterize the scenario with the highest ecological passage cost; This is a nonlinear adjustment coefficient used to enhance the resistance amplification effect in high ecological risk areas; The formula for calculating the identification of ecological corridors is as follows: ; in, As an ecological source To grid cell The minimum cumulative resistance distance; As an ecological source To grid cell The minimum cumulative resistance distance; For grid cells The ecological resistance value; As an ecological source With ecological source The total cost of the path with the least resistance between them; This is a path cost tolerance used to control the width range of ecological corridors; S3.2 Critical Path Identification: Combining node centrality, network topology, and ecological function attributes, the intensity of risky flows in the network is assessed. The system is tiered and classified to extract the controlling pathway, risk diffusion pathway, and critical connectivity pathway to identify key ecological pathways that play an important role in the structure and function of the ecosystem. The calculation method is as follows: ; in, For path Average resistance, For path average risk, As the source The least cost path, For the number of path pixels, For pixel drag, For pixel risk; ; in, Connecting edges The intensity of the risk flow, The overall importance of the two nodes is determined by their combined importance. For path average risk, For the average resistance of the path, It is a very small constant; S3.3, Comprehensive Importance Evaluation Model: Mingjiang Structural Centrality Node Ecosystem Service Function Level Average risk level of nodes Integrating and constructing a comprehensive importance index

[0017] The comprehensive model expression is: ; in, For nodes The overall importance index, For degree centrality, For compactness centrality, For the sake of mediumness centrality, For eigenvector centrality, For ecosystem service function levels, Average risk for nodes These are the weighting coefficients.

[0018] This indicator can quantitatively evaluate the comprehensive contribution of nodes to the stability of ecosystem structure and the process of risk transmission, and is an important basis for identifying key nodes.

[0019] Furthermore, the specific process of S4 is as follows: S4.1 Integrated Ecosystem Assessment: A comprehensive assessment system is constructed from three aspects: risk, structure, and ecosystem service function. The risk dimension is used to identify degradation hotspots and potential risk diffusion directions. The structure dimension reflects the overall stability of the ecosystem through network connectivity, the number of key nodes, and corridor integrity. The function dimension identifies vulnerable areas with high function but high risk based on the node service level. The three dimensions together form a systematic ecological assessment result. S4.2, Repair Priority Determination: Based on ERI and A combination of risk thresholds and importance is used to construct a remediation priority matrix; the formula for calculating the risk threshold is: ; in, The high-risk threshold The threshold for medium risk. This represents the risk value for all nodes. calculate Quantiles For nodes The average risk; the importance calculation formula is: ; in, For high importance threshold, The importance threshold is... Indicates importance to all nodes calculate Quantiles For nodes Overall importance; High-risk and high-importance nodes are designated as Level 1 remediation zones, and the criteria for their designation are as follows: High-risk but moderately important areas are classified as secondary remediation zones, defined by the following criteria: Areas with medium risk but high importance are classified as Level III remediation zones, defined by the following criteria: Low-risk, low-importance areas can be designated as general management areas, and their definition criteria are as follows: ; S4.3 Evaluation of Restoration Effect: The restoration effect is quantitatively evaluated through multi-temporal ERI and changes in ecological network structure; the degree of risk improvement is characterized by the difference in ERI before and after; the degree of structural improvement reflects the improvement of ecological connectivity through changes in network density, connectivity, and intermediation; the migration of key node levels is used to identify the trend of structural enhancement or deterioration.

[0020] This assessment system can dynamically monitor the effectiveness of restoration and guide subsequent management.

[0021] This invention constructs a framework for identifying regional ecological risks and key nodes based on multi-source remote sensing imagery, ecological and environmental indicators, and ecological network theory. It utilizes remote sensing big data and real-time computation to construct an Ecological Risk Index (ERI), enabling real-time dynamic monitoring of ecological status. Through long-term, dense ecological risk data, it constructs an ecological spatial network using ecological risk patches as units, rapidly identifying key nodes in the ecosystem using degree centrality (DC). Critical pathways are obtained based on the direct flow matrix and integrated flow matrix between nodes. Furthermore, it constructs an ecological comprehensive evaluation system, including modules for ecological spatial zoning, comprehensive ecological diagnosis, and ecological restoration assessment. By periodically updating ecological risk data and node information, it achieves dynamic closed-loop evaluation and ecological management. This invention enables rapid and accurate identification of key nodes in arid region ecosystems while simultaneously obtaining key ecological pathways. It also achieves restoration zone identification and dynamic closed-loop updates, improving the monitoring efficiency and management capabilities of arid region ecosystems. It possesses advantages such as intelligence, quantifiability, and strong adaptability, making it suitable for applications such as large-scale ecological risk monitoring, key node identification, ecological security pattern construction, and major ecological engineering planning in arid regions. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention; Figure 2 This is a flowchart of the ecosystem comprehensive evaluation and restoration effect assessment of the present invention. Detailed Implementation

[0023] The invention will now be further described with reference to the accompanying drawings.

[0024] like Figure 1 As shown, a method for identifying and evaluating key nodes in an ecosystem in arid northern regions is presented, with the following steps: S1. Ecological Risk Identification: Based on Google Earth Engine (GEE), long-term remote sensing data is called, and ecological indicators are extracted from the called remote sensing data to construct an ecological risk index, so as to realize the quantitative inversion of the spatial pattern of ecological risk in large-scale arid areas. S2. Key Ecological Node Identification: Based on the ecological spatial network analysis method, the degree centrality DC is used as the node importance evaluation index to quantitatively evaluate the connectivity of network nodes, and the nodes with DC in the top 10% are selected as the main key ecological nodes. S3. Critical Path Identification: Based on the direct flow matrix and comprehensive flow matrix between nodes, the intensity of risk flows is classified, and the main control path, risk diffusion path and key connection path in the network are extracted to obtain the critical ecological path. S4. Comprehensive Evaluation: Based on ecological risks, key nodes and key paths, construct an ecological comprehensive evaluation system, identify ecological zones, and realize dynamic closed-loop evaluation of the ecosystem.

[0025] As a preferred implementation method, such as Figure 2 As shown, the specific process of S1 is as follows: S1.1 Data Selection: To construct a spatially consistent, temporally coherent, and quality-controllable basic remote sensing dataset, providing data support for ERI construction and key ecological node identification, specifically: Medium-to-high resolution multi-source satellite remote sensing imagery acquired online through a remote sensing data cloud platform is used as the primary data foundation. The Landsat series remote sensing data released by the U.S. Geological Survey (USGS) is preferred, as it has a 30m spatial resolution and includes systematic radiometric calibration, geometric correction, and cloud masking information, making it suitable for observation environments with large variations in surface reflectance and unstable cloud cover in arid regions. To enhance the seasonal representativeness of ecological processes in arid regions, a three-month period during a typical season is used as the temporal observation window (spring: (March to May; summer: June to August; autumn: September to November; winter: December to February) to obtain stable vegetation, habitat, and thermal information. To obtain stable statistical samples for model training, areas with vegetation cover, bare ground, water bodies, and impermeable surfaces with cloud cover ≤10% were selected within the study area as training regions. The training region was constructed with the geometric center of the study area as the base point, and a buffer zone of tens of kilometers was built outwards, intersecting with the boundary of the study area to form the final training region. This training region was used to calculate subsequent quantile normalization parameters, image matching parameters, and statistical features such as the PCA covariance matrix, thereby improving model robustness and reducing the impact of missing data in marginal areas. S1.2 Remote sensing image processing: S1.2-1. To address the difficulty of directly comparing multi-temporal remote sensing images in arid regions under complex lighting, atmospheric aerosol, and sensor noise conditions, a homogenization preprocessing method for multi-source, multi-temporal remote sensing images is proposed for ecological risk assessment. This method is used to construct a basic dataset with temporal comparability and spatial consistency. The method utilizes the features included with the Landsat series products. In the quality control band, the bits corresponding to clouds, cloud shadows, and snow cover are analyzed to construct a logical mask of "non-cloud—non-cloud shadow—non-snow," automatically removing contaminated pixels. To obtain accurate surface reflectance information, the remote sensing image is radiometrically calibrated to eliminate the influence of sensor response differences. A quantile linear matching strategy based on the seasonal median benchmark is used to obtain accurate surface reflectance information. The remote sensing image is radiometrically calibrated to eliminate the influence of sensor response differences; the formula is: ; in, The pixel value of the seasonal reference image in band b For the first The pixel value of the scene image in band b. This represents the number of images within the same seasonal time window. It is a median composition operator; Using the median composite results of multiple images of the same season in the study area as a radiometric reference, the difference between the 2% and 98% quantiles of the image to be corrected and the reference image in the same band is calculated. A linear transformation function is constructed and applied to the whole image; thereby adaptively reducing the systematic radiometric bias caused by seasonal atmospheric transparency fluctuations, changes in observation angle and wind and sand environment in arid areas.

[0026] S1.2-2, A stripe suppression algorithm based on vertical direction decomposition extracts low-frequency background by applying a local smoothing kernel along the stripe direction. The original image is then decomposed into low-frequency and high-frequency components and reconstructed according to a preset ratio. This effectively reduces stripe and ripple noise while preserving surface texture and edge information, significantly improving the spatial consistency of multiple images. The calculation formula is as follows: ; in, Low-frequency background components This is the corrected image before band suppression. A one-dimensional smooth kernel along the strip direction. This represents the convolution / moving average operation; in, To smooth the window length; ; in, These are high-frequency components; ; in, The output image after stripe suppression. This is the band suppression coefficient. The larger the value, the stronger the inhibition. S1.2-3. Perform seasonal median composite analysis on the images that have undergone the above processing within a preset time window. The calculation formula is as follows: ; in, For seasonal composite images in band The pixel value, For the first Image after landscape band suppression This represents the number of valid images within the time window. It is a median composition operator; ; in, For pixels Number of valid observations within the window The minimum effective observation threshold; By eliminating short-term weather disturbances and transient sensor noise, standardized synthetic images that can stably characterize the seasonal ecological state of the study area are obtained. These images are then used as a unified input for risk index calculation, principal component analysis weight extraction, and ecological risk index (ERI) construction, thus achieving automated and homogenized conversion from raw multi-temporal images to a set of ecological risk factors.

[0027] S1.3 Risk Indicator Calculation: Risk indicators include risk receptor indicators, risk source indicators, and exposure response indicators. The calculation of each indicator is as follows: S1.3-1. Risk receptor indicators are used to characterize the health status of an ecosystem and its sensitivity to external stresses. Two types of indicators are selected: NDVI and WET. NDVI, based on red and near-infrared reflectance, accurately reflects vegetation cover and photosynthetic activity, and is the most sensitive indicator of vegetation degradation and recovery in arid regions. Median composites of seasonal images can stably present the overall vegetation status and avoid short-term weather disturbances. The calculation formula is as follows: ; in, For pixels Normalized differential vegetation index, Near-infrared reflectance, Red light reflectance, It is a very small constant; ; in, For seasonal NDVI, For the first NDVI, Median synthesis; WET, constructed using Tasseled Cap transformation, comprehensively reflects the overall moisture content of soil and vegetation. It can sensitively capture water deficit and soil drying processes in arid regions, possessing core monitoring value in arid ecosystems where water is the primary limiting factor. The calculation formula is as follows: ; in, For TasseledCap humidity component, Let reflectivity vector be the vector. This represents the humidity component coefficient vector corresponding to the sensor. For the first Each band coefficient For the first Reflectivity of each band; ; in, WET is a seasonal scale, and the first WET is the second WET. JingWET, Median synthesis; S1.3-2. Risk source indicators are used to characterize the main negative driving forces leading to ecological degradation. The SI and BSI indicators are used together to reflect the core driving factors of land quality deterioration in arid areas. The SI utilizes the spectral differences between visible light and short-wave infrared to enhance salinity information, effectively identifying salinization areas caused by salt accumulation, salinization return, or improper irrigation. The calculation formula is as follows: ; in, The salt enhancement index, For shortwave infrared reflectivity, Near-infrared reflectance, It is a very small constant; The Baseline Indicator (BSI) is constructed using blue light, red light, near-infrared light, and short-wave infrared light to reflect the degree of bare land exposure. It is an important indicator for identifying the dynamics of soil desertification, wind erosion diffusion, and degradation edges. The calculation formula is as follows: ; in, The bare soil index, For shortwave infrared reflectivity, Red light reflectance, Near-infrared reflectance, Blue light reflectance, It is a very small constant; S1.3-3. Exposure response indicators characterize the immediate response of ecosystems to environmental stresses such as water deficit and thermal stress. The LST and ET indicators are selected. LST is retrieved using thermal infrared band inversion, revealing the thermal risks caused by increased bare surface area, decreased vegetation cooling capacity, and limited surface evaporation. The calculation formula is as follows: ; in, Surface temperature, unit: Brightness temperature, unit: The center wavelength of thermal infrared, Here, h is the constant term, and h is Planck's constant. At the speed of light, Boltzmann's constant; For pixels The surface emissivity, The conversion constant; The ET index is constructed using the inverse relationship between NDVI and LST to reflect vegetation water use capacity and ecosystem energy balance. The calculation formula is as follows: ; in, As an evapotranspiration response index, For the normalized positive NDVI term, The normalized LST is used; when ET decreases, it indicates a decline in system evapotranspiration and weakening of ecological functions, which is an important response signal to drought stress. S1.4 Risk Factor Normalization: Due to the significant differences in the dimensions and numerical ranges of different indicators, a quantile-robust normalization method is adopted to ensure the comparability of each indicator in subsequent PCA and comprehensive risk assessment, and to reduce the impact of extreme values ​​and local anomalies. Details are as follows: S1.4-1. To ensure consistency in the direction of indicators, the directions of indicators are unified according to their ecological significance: the larger the values ​​of NDVI, WET, and ET, the better the ecology, so they are taken as negative values, so the larger the values, the higher the risk; the larger the values ​​of SI, BSI, and LST, the higher the risk, so they are kept in their original directions. S1.4-2. Calculate the 2nd and 98th quantiles for each indicator within the training region to construct a robust interval; based on this interval, normalize the indicator values ​​to 0~1 in the following manner: ; in, As an indicator The normalization result, and Indicators The training region is divided into 2% and 98% quantiles, and the data is limited to the effective range to prevent normalization distortion due to local anomalies or noise. When the quantile difference is too small or data is missing, minimum / maximum values ​​are used for backtracking normalization to ensure algorithm stability. The calculation formula is as follows: ; in, To revert to the normalization result, The index value after unifying the direction. These are the minimum and maximum values, respectively. It is a very small constant; after normalization, all six categories of indicators follow similar numerical ranges and have the same direction, providing a unified input for subsequent PCA weighting and ERI construction; S1.5, PCA Weight Extraction and Eigenvector Direction Constraint: The PCA method is used to extract the comprehensive risk gradient of six types of indicators, and a weight system for the ecological risk index is constructed, specifically as follows: S1.5-1. Construct the normalized indicators as an array, using pixels within the training region as samples. Obtain the correlation structure between indicators by calculating the sample covariance matrix. The calculation formula is as follows: ; in, The sample covariance matrix is ​​used to characterize the correlation structure among indicators. The centered sample matrix, This represents the number of training samples; To improve statistical robustness, a two-level strategy is adopted to estimate the sample covariance: when the number of training samples is sufficient and the covariance can be calculated, the sample covariance matrix is ​​used; when the number of samples is insufficient or the structure is abnormal, the calculation regresses to the region covariance matrix to ensure matrix invertibility and the safety of eigenvalue decomposition. The formula for the regression mechanism is as follows: ; in, This is the safety matrix used for eigenvalue decomposition. The region covariance matrix, The diagonal regularization coefficient is... It is the identity matrix; S1.5-2. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and eigenvectors of each principal component; select PC1 as the comprehensive risk gradient, whose eigenvector elements reflect the contribution of each indicator in the risk direction. The formulas for eigenvalue decomposition and PC1 are as follows: ; in, For the first 1 eigenvalue, For the corresponding feature vector, Principal component number, It is the covariance matrix; ; in,, The variance explained by the first principal component, PC1. The largest eigenvalue, It is the sum of all eigenvalues; ; in, For pixels PC1 score, The eigenvector of PC1 is the first One element, A normalized index value; To ensure consistency in ecological significance, according to The weight sign of the vegetation degradation index imposes a directional constraint on the entire feature vector: when When the weights of the eigenvectors in the first principal component are negative, the overall eigenvector is multiplied by -1, making... The contribution is positive, ensuring that the larger the PC1 value, the higher the ecological risk. This yields the PCA weights corresponding to the six indicators, realizing a data-driven and consistent ecological risk weight system. S1.6 Construction of ERI: After obtaining the PCA weighting system, the six normalized risk indicators are weighted and superimposed according to their weights to obtain a preliminary ecological risk index: ; in, For PCA weights, To normalize the index value, and to enhance numerical stability, the ERI undergoes two-stage normalization: training area normalization, which standardizes the value using quantiles or extreme value ranges within the training area, ensuring it roughly conforms to the normal distribution within that area. The range is calculated using the following formula: ; in, The normalized risk index for the training area. Training areas The 2% and 98th percentiles, It is a very small constant; the study area is normalized as a whole: then it is normalized again according to the statistical interval of the entire study area, so that the final ERI is uniformly distributed within the study area. The interval is calculated using the following formula: ; in, This represents the final ecological risk index after normalization for the entire study area. The 2% and 98th percentiles, It is a very small constant; the higher the ERI value, the greater the risk to the ecosystem, and it is used to reflect the comprehensive ecological risk level of multiple factors such as vegetation degradation, soil salinization, bare land expansion, and high temperature stress. S1.7 Construction of the Ecological Risk Spatial Network: After obtaining the spatial distribution of ERI, the network is configured to divide the study area into spatial nodes using a fixed grid of 30m × 30m, and each node is assigned attributes such as area, ecological type, ecosystem service level, and mean ERI. Edges are constructed based on the spatial relationships between nodes, generated through the minimum cumulative resistance (MCR) path based on the ecological resistance surface. Ecological resistance is provided by risk source indicators BSI, SI, and LST; higher risk results in greater resistance, while risk receptors and exposure response indicators can reduce resistance. The formula for constructing the resistance surface is: ; in, This represents the pixel resistance value. It is the minimum ecological resistance constant. As a normalized indicator for risk source categories, For the normalized index of the sustained-release term, As the risk source weight, To mitigate the weight, .

[0028] Step S1 can be used to construct a spatial connectivity network that conforms to the characteristics of ecological processes in arid regions, in order to identify nodes that play a key regulatory role in the ecological structure.

[0029] As a preferred embodiment, the specific process of S2 is as follows: S2.1 Calculate the node centrality index based on the ecological space network, and use the natural breakpoint method to... To classify and form Level of importance nodes; Level nodes are core ecological control nodes and are crucial for maintaining regional ecological stability; Level nodes are important support nodes, and are mostly located in corridor intersection areas; Level nodes are general ecological nodes; Level 1 nodes are mostly located in the structural edge area or high-risk degradation nodes, and therefore have priority for repair. S2.2. Based on node level, the study area is divided into ecological protection priority zones, i.e. Level nodes and their affected areas, corridor control areas, i.e. Level nodes and key corridors, key restoration areas Level nodes and their affected areas and general management areas Level nodes and the remaining areas.

[0030] Furthermore, the specific process of S3 is as follows: S3.1 Network Edge Construction: Based on ERI, an ecological resistance surface is constructed, and ecological corridors are identified through MCR paths to achieve a realistic simulation of potential connectivity structures. The uneven resistance in arid regions and the diffusion of ecological processes along corridors are fully considered. The formula for calculating the ecological resistance surface is as follows: ; in, For grid cells Ecological resistance value at the location; For grid cells The ecological risk index after quantile normalization; This is the minimum ecological resistance constant, used to characterize the case where ecological passage costs are minimized; This is the maximum ecological resistance constant, used to characterize the scenario with the highest ecological passage cost; This is a nonlinear adjustment coefficient used to enhance the resistance amplification effect in high ecological risk areas; The formula for calculating the identification of ecological corridors is as follows: ; ; ; in, As an ecological source To grid cell The minimum cumulative resistance distance; As an ecological source To grid cell The minimum cumulative resistance distance; For grid cells The ecological resistance value; As an ecological source With ecological source The total cost of the path with the least resistance between them; This is a path cost tolerance used to control the width range of ecological corridors; S3.2 Critical Path Identification: Combining node centrality, network topology, and ecological function attributes, the intensity of risky flows in the network is assessed. The system is tiered and classified to extract the controlling pathway, risk diffusion pathway, and critical connectivity pathway to identify key ecological pathways that play an important role in the structure and function of the ecosystem. The calculation method is as follows: ; in, For path Average resistance, For path average risk, As the source The least cost path, For the number of path pixels, For pixel drag, For pixel risk; ; in, Connecting edges The intensity of the risk flow, The overall importance of the two nodes is determined by their combined importance. For path average risk, For the average resistance of the path, It is a very small constant; S3.3, Comprehensive Importance Evaluation Model: Mingjiang Structural Centrality Node ecosystem service function level S and node average risk level Integrating and constructing a comprehensive importance index The comprehensive model expression is: ; in, For nodes The overall importance index, For degree centrality, For compactness centrality, For the sake of mediumness centrality, For eigenvector centrality, For ecosystem service function levels, Average risk for nodes These are the weighting coefficients.

[0031] This indicator can quantitatively evaluate the comprehensive contribution of nodes to the stability of ecosystem structure and the process of risk transmission, and is an important basis for identifying key nodes.

[0032] As a preferred embodiment, the specific process of S4 is as follows: S4.1 Integrated Ecosystem Assessment: A comprehensive assessment system is constructed from three aspects: risk, structure, and ecosystem service function. The risk dimension is used to identify degradation hotspots and potential risk diffusion directions. The structure dimension reflects the overall stability of the ecosystem through network connectivity, the number of key nodes, and corridor integrity. The function dimension identifies vulnerable areas with high function but high risk based on the node service level. The three dimensions together form a systematic ecological assessment result. S4.2, Determining Repair Priority: Based on A combination of risk thresholds and importance is used to construct a remediation priority matrix; the formula for calculating the risk threshold is: ; in, The high-risk threshold The threshold for medium risk. This represents the risk value for all nodes.

[0033] calculate Quantiles For nodes The average risk; the importance calculation formula is: ; in, For high importance threshold, The importance threshold is... Indicates importance to all nodes

[0034] calculate Quantiles For nodes Overall importance; High-risk and high-importance nodes are designated as Level 1 remediation zones, and the criteria for their designation are as follows: High-risk but moderately important areas are classified as secondary remediation zones, defined by the following criteria: Areas with medium risk but high importance are classified as Level III remediation zones, defined by the following criteria: Low-risk, low-importance areas can be designated as general management areas, and their definition criteria are as follows: ; S4.3 Evaluation of Restoration Effect: The restoration effect is quantitatively evaluated through multi-temporal ERI and changes in ecological network structure; the degree of risk improvement is characterized by the difference in ERI before and after; the degree of structural improvement reflects the improvement of ecological connectivity through changes in network density, connectivity, and intermediation; the migration of key node levels is used to identify the trend of structural enhancement or deterioration.

[0035] This assessment system can dynamically monitor the effectiveness of restoration and guide subsequent management.

Claims

1. A method for identifying and evaluating key nodes in an ecosystem in arid northern regions, characterized in that, The steps are as follows: S1. Ecological Risk Identification: Based on Google Earth Engine (GEE), long-term remote sensing data is called, and ecological indicators are extracted from the called remote sensing data to construct an ecological risk index, so as to realize the quantitative inversion of the spatial pattern of ecological risk in large-scale arid areas. S2. Key Ecological Node Identification: Based on the ecological spatial network analysis method, the degree centrality DC is used as the node importance evaluation index to quantitatively evaluate the connectivity of network nodes, and the nodes with DC in the top 10% are selected as key ecological nodes. S3. Critical Path Identification: Based on the direct flow matrix and comprehensive flow matrix between nodes, the intensity of risk flows is classified, and the main control path, risk diffusion path and key connection path in the network are extracted to obtain the critical ecological path. S4. Comprehensive Evaluation: Based on ecological risks, key nodes and key paths, construct an ecological comprehensive evaluation system, identify ecological zones, and realize dynamic closed-loop evaluation of the ecosystem; The specific process of S2 is as follows: S2.1 Calculate the node centrality index based on the ecological space network, and use the natural breakpoint method to calculate the comprehensive importance index of node i. To classify and form Level of importance nodes; Level nodes are core ecological control nodes and are crucial for maintaining regional ecological stability; Level nodes are important support nodes, and are mostly located in corridor intersection areas; Level nodes are general ecological nodes; Level 1 nodes are mostly located in the structural edge area or high-risk degradation nodes, and therefore have priority for repair. S2.

2. Based on node level, the study area is divided into ecological protection priority zones, i.e. Level nodes and their affected areas, corridor control areas, i.e. Level nodes and key corridors, key restoration areas Level nodes and their affected areas and general management areas Level nodes and the remaining areas; The specific process of S3 is as follows: S3.1 Network Edge Construction: Based on ERI, an ecological resistance surface is constructed, and ecological corridors are identified through MCR paths. The uneven resistance in arid regions and the diffusion of ecological processes along corridors are fully considered. The formula for calculating the ecological resistance surface is as follows: ； in, For grid cells Ecological resistance value at the location; For grid cells The ecological risk index after quantile normalization; This is the minimum ecological resistance constant, used to characterize the case where ecological passage costs are minimized; This is the maximum ecological resistance constant, used to characterize the scenario with the highest ecological passage cost; This is a nonlinear adjustment coefficient used to enhance the resistance amplification effect in high ecological risk areas; The formula for calculating the identification of ecological corridors is as follows: ； ； ； in, As an ecological source To grid cell The minimum cumulative resistance distance; As an ecological source To grid cell The minimum cumulative resistance distance; For grid cells The ecological resistance value; As an ecological source With ecological source The total cost of the path with the least resistance between them; This is a path cost tolerance used to control the width range of ecological corridors; S3.2 Critical Path Identification: Combining node centrality, network topology, and ecological function attributes, the intensity of risky flows in the network is assessed. The system is tiered and classified to extract the controlling pathway, risk diffusion pathway, and critical connectivity pathway to identify key ecological pathways that play an important role in the structure and function of the ecosystem. The calculation method is as follows: ； in, For path Average resistance, For path average risk, As the source The least cost path, For the number of path pixels, For pixel drag, For pixel risk; ； in, Connecting edges The intensity of the risk flow, The overall importance of the two nodes is determined by their combined importance. For path average risk, For the average resistance of the path, It is a very small constant; S3.3, Comprehensive Importance Evaluation Model: Mingjiang Structural Centrality Node ecosystem service function level S and node average risk level Integrating and constructing a comprehensive importance index The comprehensive model expression is: ； in, For nodes The overall importance index, For degree centrality, For compactness centrality, For the sake of mediumness centrality, For eigenvector centrality, For ecosystem service function levels, Average risk for nodes These are the weighting coefficients.

2. The method for identifying and evaluating key nodes in the ecosystem of arid northern regions according to claim 1, characterized in that, The specific process of S1 is as follows: S1.1 Data Selection: Construct a spatially consistent, temporally coherent, and quality-controllable basic remote sensing dataset to provide data support for ERI construction and key ecological node identification. Specifically, Landsat series remote sensing images with cloud cover ≤10% are selected online using GEE as the data basis. Three consecutive months of typical seasons are used as the time observation window. Within the study area, vegetation cover, bare surface, water body, and impermeable surface areas are selected as training areas. The training area is constructed with the geometric center of the study area as the base point, and a buffer zone is built outward from it. The final training area is formed after intersecting with the boundary of the study area. S1.2 Remote sensing image processing: S1.2-1. Using the quality control band, the bits corresponding to clouds, cloud shadows, and snow cover are analyzed to construct a logical mask of "non-cloud—non-cloud shadow—non-snow" to automatically remove contaminated pixels. Subsequently, based on a quantile linear matching strategy using the seasonal median benchmark, the true surface reflectance information is obtained, and the remote sensing image is radiometrically calibrated to eliminate the influence of sensor response differences. The formula is: ； in, Seasonal reference image in band The pixel value, For the first Scene image in band The pixel value, This represents the number of images within the same seasonal time window. It is a median composition operator; Using the median composite of multiple images of the same season in the study area as a radiometric reference, the radiometric values ​​of the image to be corrected and the reference image are calculated in the same band. and The difference in quantiles is used to construct a linear transformation function and apply it to the entire scene image; S1.2-2, A stripe suppression algorithm based on vertical decomposition extracts low-frequency background by applying a local smoothing kernel along the stripe direction, and then reconstructs the original image by decomposing it into low-frequency and high-frequency components according to a preset ratio. The calculation formula is as follows: ； in, For low-frequency background components, This is the corrected image before band suppression. A one-dimensional smooth kernel along the strip direction. This represents the convolution / moving average operation; ； in, To smooth the window length; ； in, These are high-frequency components; ； in, The output image after stripe suppression. This is the band suppression coefficient. The larger the value, the stronger the inhibition. S1.2-3. Perform seasonal median composite analysis on the images that have undergone the above processing within a preset time window. The calculation formula is as follows: ； in, For seasonal composite images in band The pixel value, For the first Image after landscape band suppression This represents the number of valid images within the time window. It is a median composition operator; ； in, For pixels Number of valid observations within the window The minimum effective observation threshold; S1.3 Risk Indicator Calculation: Risk indicators include risk receptor indicators, risk source indicators, and exposure response indicators. The calculation of each indicator is as follows: S1.3-1. Risk receptor indicators are used to characterize the health status of an ecosystem and its sensitivity to external stresses. Two types of indicators are selected: NDVI and WET. NDVI is constructed based on red light and near-infrared reflectance, and the calculation formula is as follows: ； in, For pixels Normalized differential vegetation index, Near-infrared reflectance, Red light reflectance, It is a very small constant; ； in, For seasonal NDVI, For the first NDVI, Median synthesis; WET is constructed using Tasseled Cap transformation to comprehensively reflect the overall moisture content of soil and vegetation. The calculation formula is as follows: ； in, For TasseledCap humidity component, Let reflectivity vector be the vector. This represents the humidity component coefficient vector corresponding to the sensor. For the first Each band coefficient For the first Reflectivity of each band; ； in, WET on a seasonal scale For the first JingWET, Median synthesis; S1.3-2. Risk source indicators are used to characterize the negative driving forces leading to ecological degradation. The SI and BSI indicators are used together to reflect the core driving factors of land quality deterioration in arid areas. The SI utilizes the spectral differences between visible light and short-wave infrared to enhance salinity information; the calculation formula is as follows: ； in, The salt enhancement index, For shortwave infrared reflectivity, Near-infrared reflectance, It is a very small constant; The Baseline Indicator (BSI) is constructed using blue light, red light, near-infrared light, and short-wave infrared light to reflect the degree of bare land exposure. It is an important indicator for identifying the dynamics of soil desertification, wind erosion diffusion, and degradation edges. The calculation formula is as follows: ; in, The bare soil index, For shortwave infrared reflectivity, Red light reflectance, Near-infrared reflectance, Blue light reflectance, It is a very small constant; S1.3-3. Exposure response indicators characterize the immediate response of ecosystems to water deficit and thermal stress. LST and ET indicators are selected. LST is retrieved using thermal infrared spectroscopy to reveal the thermal risks caused by increased bare surface area, decreased vegetation cooling capacity, and limited surface evaporation. The calculation formula is as follows: ； in, Surface temperature, unit: , Brightness temperature, unit: , The center wavelength of thermal infrared, For constant terms, Let be Planck's constant. At the speed of light, Boltzmann's constant; For pixels The surface emissivity, The conversion constant; The ET index is constructed using the inverse relationship between NDVI and LST to reflect vegetation water use capacity and ecosystem energy balance. The calculation formula is as follows: ； in, As an evapotranspiration response index, For the normalized positive NDVI term, The normalized LST is used; when ET decreases, it indicates a decline in system evapotranspiration and weakening of ecological functions, which is an important response signal to drought stress. S1.4 Risk Factor Normalization: A quantile-robust normalization method is adopted, as follows: S1.4-1. To ensure consistency in the direction of indicators, the directions of indicators are unified according to their ecological significance: the larger the values ​​of NDVI, WET, and ET, the better the ecology, so they are taken as negative values, so the larger the values, the higher the risk; the larger the values ​​of SI, BSI, and LST, the higher the risk, so they are kept in their original directions. S1.4-2. Calculate the 2nd and 98th quantiles for each indicator within the training region to construct a robust interval; based on this interval, normalize the indicator values ​​to the following method. : ； in, The normalization result, Indicators In the training region, at the 2% and 98% quantiles, when the quantile difference is too small or data is missing, a minimum / maximum value is used for backtracking normalization to ensure algorithm stability. The calculation formula is as follows: ； in, To revert to the normalization result, The index value after unifying the direction. These are the minimum and maximum values, respectively. It is a very small constant; after normalization, all six categories of indicators follow similar numerical ranges and have the same direction, providing a unified input for subsequent PCA weighting and ERI construction; S1.5, PCA Weight Extraction and Eigenvector Direction Constraint: The PCA method is used to extract the comprehensive risk gradient of six types of indicators, and a weight system for the ecological risk index is constructed, specifically as follows: S1.5-1. Construct the normalized indicators as an array, using pixels within the training region as samples. Obtain the correlation structure between indicators by calculating the sample covariance matrix. The calculation formula is as follows: ； in, The sample covariance matrix is ​​used to characterize the correlation structure among indicators. The centered sample matrix, This represents the number of training samples; To improve statistical robustness, a two-level strategy is adopted to estimate the sample covariance: when the number of training samples is sufficient and the covariance can be calculated, the sample covariance matrix is ​​used; when the number of samples is insufficient or the structure is abnormal, the calculation regresses to the region covariance matrix to ensure matrix invertibility and the safety of eigenvalue decomposition. The formula for the regression mechanism is as follows: ； in, This is the safety matrix used for eigenvalue decomposition. The region covariance matrix, The diagonal regularization coefficient is... It is the identity matrix; S1.5-2. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and eigenvectors of each principal component; select PC1 as the comprehensive risk gradient, whose eigenvector elements reflect the contribution of each indicator in the risk direction. The formulas for eigenvalue decomposition and PC1 are as follows: ； in, For the first 1 eigenvalue, For the corresponding feature vector, Principal component number, It is the covariance matrix; ； in, The variance explained by the first principal component, PC1. The largest eigenvalue, It is the sum of all eigenvalues; ； in, For pixels PC1 score, The eigenvector of PC1 is the first One element, A normalized index value; To ensure consistency in ecological significance, according to The weight sign of the vegetation degradation index imposes a directional constraint on the entire feature vector: when When the weights of the eigenvectors in the first principal component are negative, the overall eigenvector is multiplied by -1, making... The contribution is positive, ensuring that the larger the PC1 value, the higher the ecological risk. This yields the PCA weights corresponding to the six indicators, realizing a data-driven and consistent ecological risk weight system. S1.6 Construction of ERI: After obtaining the PCA weighting system, the six normalized risk indicators are weighted and superimposed according to their weights to obtain a preliminary ecological risk index: ； in, For PCA weights, To normalize the index value, and to enhance numerical stability, the ERI undergoes two-stage normalization: training area normalization, which standardizes the value using quantiles or extreme value ranges within the training area, ensuring it roughly conforms to the normal distribution within that area. The range is calculated using the following formula: ； in, The normalized risk index for the training area. Training areas The 2% and 98th percentiles, It is a very small constant; the study area is normalized as a whole: then it is normalized again according to the statistical interval of the entire study area, so that the final ERI is uniformly distributed within the study area. The interval is calculated using the following formula: ； in, This represents the final ecological risk index after normalization for the entire study area. The 2% and 98th percentiles, It is a very small constant; the higher the ERI value, the greater the risk to the ecosystem, and it is used to reflect the comprehensive ecological risk level of multiple factors such as vegetation degradation, soil salinization, bare land expansion, and high temperature stress. S1.7 Construction of the Ecological Risk Spatial Network: After obtaining the spatial distribution of ERI, it is configured to divide the study area into spatial nodes according to a fixed grid, and each node is assigned the attributes of area, ecological type, ecosystem service level, and mean ERI. Edges are constructed based on the spatial relationships between nodes, generated through the minimum cumulative resistance (MCR) path based on the ecological resistance surface. Ecological resistance is provided by risk source indicators BSI, SI, and LST; the higher the risk, the greater the resistance. Risk receptors and exposure response indicators can reduce resistance. The formula for constructing the resistance surface is: ； in, This represents the pixel resistance value. It is the minimum ecological resistance constant. As a normalized indicator for risk source categories, For the normalized index of the sustained-release term, As the risk source weight, To mitigate the weight, .

3. The method for identifying and evaluating key nodes in the ecosystem of arid northern regions according to claim 1, characterized in that, The specific process of S4 is as follows: S4.1 Integrated Ecosystem Assessment: A comprehensive assessment system is constructed from three aspects: risk, structure, and ecosystem service function. The risk dimension is used to identify degradation hotspots and potential risk diffusion directions. The structure dimension reflects the overall stability of the ecosystem through network connectivity, the number of key nodes, and corridor integrity. The function dimension identifies vulnerable areas with high function but high risk based on the node service level. The three dimensions together form a systematic ecological assessment result. S4.2, Determining Repair Priority: Based on A combination of risk thresholds and importance is used to construct a remediation priority matrix; the formula for calculating the risk threshold is: ； in, The high-risk threshold The threshold for medium risk. This represents the risk value for all nodes. ; calculate Quantiles For nodes The average risk; the importance calculation formula is: ； in, For high importance threshold, The importance threshold is... Indicates importance to all nodes ; calculate Quantiles For nodes Overall importance; High-risk and high-importance nodes are designated as Level 1 remediation zones, and the criteria for their designation are as follows: High-risk but moderately important areas are classified as secondary remediation zones, defined by the following criteria: Areas with medium risk but high importance are classified as Level III remediation zones, defined by the following criteria: Low-risk, low-importance areas can be designated as general management areas, and their definition criteria are as follows: ; S4.3 Evaluation of Restoration Effect: The restoration effect is quantitatively evaluated through multi-temporal ERI and changes in ecological network structure; the degree of risk improvement is characterized by the difference in ERI before and after; the degree of structural improvement reflects the improvement of ecological connectivity through changes in network density, connectivity, and intermediation; the migration of key node levels is used to identify the trend of structural enhancement or deterioration.

Images