A method for assessing the hazards of job collapse, analyzing the urgency of its management, and optimizing its management.

Abstract

This invention provides a method for assessing the hazard of ridge erosion, analyzing the urgency of remediation, and optimizing remediation efforts. The method includes: acquiring and preprocessing multi-source data; identifying extreme weather events and establishing a model linking them to ridge erosion; assessing the intensity of anthropogenic disturbances; extracting ridge erosion characteristics and verifying them in the field; constructing an evaluation index system by integrating extreme weather, anthropogenic disturbances, and ridge erosion characteristics; constructing a CNN-RNN-Transformer fusion deep learning model for comprehensive hazard assessment; constructing an urgency assessment model based on the evaluation results and considering factors such as the township's leading industries and the economic feasibility of remediation; using GIS technology to visualize the evaluation results, generating a grade distribution map, and proposing remediation measures; optimizing the parameters of the remediation measures; and constructing an industry-specific economic benefit model to assess the economic feasibility of remediation. This invention solves the technical problem that traditional ridge erosion hazard assessments do not fully consider external factors such as extreme weather and anthropogenic disturbances, leading to biased evaluation results and difficulty in accurate remediation.

Description

Technical Field

[0001] This invention relates to the technical field of soil and water conservation and ecological security assessment, and in particular to a method for assessing the hazards of gully collapse, analyzing the urgency of remediation, and optimizing remediation. Background Technology

[0002] Benggang is one of the most serious types of soil erosion in the red soil region of southern my country. It is characterized by high erosion intensity, wide range of damage, and high difficulty in control, seriously threatening regional ecological security, agricultural production, and socio-economic development.

[0003] In recent years, intensified global climate change has led to frequent extreme weather events such as torrential rains and typhoons, significantly inducing and exacerbating ridge erosion. Simultaneously, increasing human activity, including vegetation destruction, engineering construction, and irrational land use, further amplifies the risk of ridge erosion. Traditional methods for assessing the hazard of ridge erosion primarily focus on the morphological characteristics of the ridge itself, such as area, depth, and volume, as well as internal factors like sediment yield. They fail to fully integrate the combined effects of external factors such as extreme weather and human disturbance, resulting in significant discrepancies between assessment results and the actual severity of ridge erosion. This makes it difficult to accurately identify key areas and priorities for ridge erosion control. Furthermore, these methods lack quantitative analysis of the urgency of control, systematic optimization of control measures, and consideration of economic feasibility, failing to meet the current practical needs for precise prevention and scientific management of ridge erosion.

[0004] Therefore, there is an urgent need for a method that can solve the above-mentioned technical problems, including assessment of the hazards of job loss, analysis of the urgency of governance, and optimization of governance. Summary of the Invention

[0005] The purpose of this invention is to provide a method for assessing the hazards of gully erosion, analyzing the urgency of remediation, and optimizing remediation. This method solves the technical problem that traditional gully erosion hazard assessments do not fully consider external factors such as extreme weather and human disturbance, leading to biased assessment results and difficulty in accurate remediation.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides a method for assessing the hazard of ridge erosion, analyzing the urgency of remediation, and optimizing remediation, comprising the following steps: S1. Acquiring meteorological data, topographic data, soil data, vegetation data, anthropogenic disturbance data, and remote sensing imagery of the ridge erosion area, and preprocessing them; S2. Identifying extreme climate events in the ridge erosion area and establishing a correlation mechanism model between the extreme climate events and ridge erosion; S3. Constructing an anthropogenic disturbance intensity assessment index system to assess the anthropogenic disturbance intensity in the ridge erosion area; S4. Extracting ridge erosion features from the preprocessed remote sensing images, verifying the ridge erosion features in the field, and conducting investigation and analysis in conjunction with ridge erosion micro-landforms; the ridge erosion micro-landforms include catchment slopes, ridge walls, alluvial cones, gullies, and alluvial fans; S5. Constructing a ridge erosion hazard assessment index system based on the extreme climate events, the anthropogenic disturbance intensity, and the ridge erosion features; S6. Constructing a deep neural network integrating convolutional neural networks, recurrent neural networks, and self-attention mechanism neural networks. S7. Based on the assessment index system for the harm of hill erosion and the deep learning model, a comprehensive evaluation of the harm of hill erosion is conducted to obtain the assessment result of the harm of hill erosion; S8. Based on the assessment result of the harm of hill erosion, a hill erosion control urgency assessment model is constructed by combining multiple factors to obtain the assessment result of the urgency of hill erosion control; the multiple factors include at least the township's leading industry and the economic feasibility of control; S9. Using geographic information system technology, the assessment result of the harm of hill erosion and the assessment result of the urgency of hill erosion control are visualized to generate a grade distribution map, and targeted hill erosion control measures and suggestions are proposed based on the grade distribution map; S10. Based on the assessment result of the harm of hill erosion and spatial interpolation data, a multi-objective optimization model with control benefits and control costs as optimization objectives is constructed to optimize the parameters of the hill erosion control measures; S11. An economic benefit model for hill erosion control under different leading industry types is constructed to assess the economic feasibility of hill erosion control.

[0007] Furthermore, in S1, the meteorological data includes rainfall, rainfall intensity, temperature, and wind speed; the topographic data includes elevation, slope, and aspect; the soil data includes soil type, soil texture, and soil organic matter content; the vegetation data includes vegetation type and vegetation coverage; the anthropogenic disturbance data includes land use type and engineering construction activities; and the remote sensing imagery includes high-resolution satellite imagery with a spatial resolution of not less than 0.5 meters and drone aerial imagery.

[0008] Furthermore, in S2, the establishment of the correlation mechanism model between extreme climate events and ridge erosion includes the following steps: based on preprocessed meteorological data, extreme value theory and frequency analysis methods are used to identify extreme climate events; using the extreme climate events as input variables, a joint probability distribution model of extreme climate events and ridge erosion is constructed using a Copula function, and the Markov chain method is used to simulate the evolution process of ridge erosion under different extreme climate event states; the influence mechanism of extreme climate events on the ridge erosion process is analyzed through the joint probability distribution model, and the contribution of the frequency, intensity, and duration of extreme climate events to the amount of ridge erosion is quantified; the contribution is used to construct the ridge erosion hazard evaluation index system.

[0009] Furthermore, in S4, the characteristics of the landslide include morphological characteristics, erosion and sediment yield characteristics, and developmental stage characteristics; the morphological characteristics include area, depth, volume, slope, perimeter, and aspect; the erosion and sediment yield characteristics include erosion modulus and sediment yield; and the developmental stage characteristics include active period and stable period.

[0010] Furthermore, in S5, the construction of the evaluation index system for the hazard of ridge erosion includes the following steps: using the analytic hierarchy process (AHP) to calculate the subjective weights of the evaluation indicators; using the entropy weight method to calculate the objective weights of the evaluation indicators; calculating the combined weights of the evaluation indicators based on the subjective and objective weights; using grey relational analysis to calculate the correlation between the evaluation indicators and the hazard of ridge erosion, and removing indicators with a correlation lower than a preset threshold; using principal component analysis to extract principal component factors with a cumulative variance contribution rate greater than a preset threshold, and clustering and reorganizing the retained evaluation indicators according to the loading matrix of the principal component factors to obtain an optimized evaluation index system.

[0011] Furthermore, in S6, the construction of a deep learning model integrating convolutional neural networks, recurrent neural networks, and self-attention mechanism neural networks includes the following steps: constructing a convolutional neural network module to extract the spatial distribution and morphological features of gully erosion; constructing a recurrent neural network module to extract the temporal variation features of extreme climate events and human disturbances; constructing a neural network module based on a self-attention mechanism to extract the global correlation features between various evaluation indicators; adaptively weighting and fusing the output features of the convolutional neural network module, recurrent neural network module, and self-attention mechanism neural network module through a fusion layer, inputting them into a fully connected layer, and outputting the gully erosion hazard assessment result; training and optimizing the deep learning model using ensemble learning and transfer learning methods; and verifying the gully erosion hazard assessment result using sensitivity analysis and uncertainty analysis methods.

[0012] Furthermore, in S7, the construction of the urgency assessment model for gully erosion control includes the following steps: based on the assessment results of gully erosion hazard, frequency of extreme weather events, intensity of human disturbance, regional ecological sensitivity, regional socio-economic importance, township leading industries, and assessment factors of economic feasibility of control, a gully erosion control urgency assessment model is constructed; a judgment matrix is ​​constructed, and the subjective weight of each assessment factor is calculated using the analytic hierarchy process (AHP); the objective weight of each assessment factor is calculated using the fuzzy comprehensive evaluation method, based on the subjective weight and combined with the measured data of each assessment factor; the combined weight of the assessment factors is comprehensively determined according to the subjective weight and objective weight; the urgency of gully erosion control is divided into four levels: low, medium, high, and extremely high, and a gully erosion control priority plan is formulated based on the urgency level results.

[0013] Furthermore, in S9, the parameter optimization includes the following steps: determining the value range of decision variables for the engineering measure parameter set and the biological measure parameter set based on spatial interpolation data; constructing a multi-objective fitness function with the optimization objectives of maximizing governance benefits and minimizing governance costs; encoding the decision variables using a genetic algorithm, and obtaining a Pareto optimal solution set by iteratively solving the multi-objective fitness function; and determining the optimal parameter combination of the hill collapse governance measures based on the Pareto optimal solution set.

[0014] Furthermore, in S10, the construction of the economic benefit model for hill collapse management under different leading industry types includes the following steps: based on the urgency assessment results of hill collapse management and the optimized parameter combination of the hill collapse management measures, determine the type and scale of the hill collapse management engineering measures; obtain industrial structure data of the township where the hill collapse occurs, and determine the leading industry type; perform spatial overlay analysis of the optimized parameter combination and the leading industry type to determine the potential impact range of the management measures on different industries; calculate the total management cost according to the determined engineering measure type and scale; predict the total management benefit according to the potential impact range and industry output data; calculate dynamic economic evaluation indicators according to the total management cost and total management benefit; the dynamic economic evaluation indicators include net present value, internal rate of return, and investment payback period; and feed the dynamic economic evaluation indicators back to the multi-objective optimization model as constraints for parameter optimization.

[0015] Furthermore, the method also includes analyzing the differences in the quantitative application of the economic benefit model in different townships, including the following steps: collecting natural geographical data and socio-economic statistics of the townships where the landslides occurred, and constructing township feature vectors; inputting the township feature vectors into the multi-objective optimization model, and calculating the landslide governance costs and benefits for different townships by adjusting the cost coefficients and benefit coefficients in the multi-objective optimization model; establishing a township comparison matrix including net present value, internal rate of return, and investment payback period; identifying key influencing factors affecting economic feasibility based on the differences in the comparison matrix; and dynamically adjusting the weights of the human disturbance intensity assessment index system or adjusting the optimization parameter combination of the landslide governance measures based on the key influencing factors.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: This invention introduces extreme climate events as an independent dimension into the gully erosion assessment system. It employs a Copula function and Markov chain method to establish a joint probability distribution model of extreme climate and gully erosion, achieving a quantitative analysis of the contribution of extreme climate to gully erosion and filling the gap in the qualitative description of extreme climate factors in traditional assessments. Simultaneously, it constructs a deep learning model fused with CNN-RNN-Transformer, using convolutional neural networks to extract spatial features of gully erosion, recurrent neural networks to extract temporal variation features, and Transformer to extract global correlation features. The resulting weighted fusion outputs the assessment results, improving the accuracy and reliability of gully erosion hazard assessment.

[0017] This invention incorporates the township's leading industries and the economic feasibility of governance into the urgency analysis of governance. Based on seven factors—the severity of landslides, frequency of extreme weather, intensity of human disturbance, ecological sensitivity, socio-economic importance, township's leading industries, and economic feasibility of governance—an urgency assessment system is constructed and assigned specific weights. The urgency is divided into four levels: low, medium, high, and extremely high. This allows governance resources to be tilted towards the areas with the most urgent needs, realizing a shift from "passive governance" to "proactive prevention and control," and providing management departments with a clear basis for prioritizing governance decisions.

[0018] This invention constructs an economic benefit model for gully erosion control based on industry type, calculating the control benefits separately for different leading industries such as agriculture, forestry, and tourism. It combines engineering construction costs, vegetation restoration costs, and monitoring and maintenance costs to calculate net present value, internal rate of return, and investment payback period, achieving a quantitative assessment of the economic feasibility of control. Simultaneously, it employs multi-objective optimization algorithms and genetic algorithms to optimize the overall parameters of engineering measures such as intercepting ditches and retaining walls, taking into account control effectiveness, engineering costs, and ecological benefits. This forms a complete technical closed loop from evaluation, classification, optimization to economic assessment, providing a scientific, precise, and feasible system solution for gully erosion prevention and control. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the methods for assessing the hazards of landslides, analyzing the urgency of remediation, and optimizing remediation provided in this embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] This embodiment provides a method for assessing the hazards of landslides, analyzing the urgency of remediation, and optimizing remediation efforts. Figure 1 As shown, it covers ten steps, from data acquisition and preprocessing, extreme climate correlation modeling, human disturbance assessment, landslide feature extraction, hazard assessment index system construction, deep learning hazard assessment, multi-factor urgency assessment, GIS visualization, engineering measure parameter optimization and economic benefit model construction, forming a closed-loop technical solution from evaluation to decision-making to governance optimization.

[0024] The specific steps are as follows: S1. Acquire multi-source data such as meteorological, topographic, soil, vegetation, and human disturbance data, as well as high-resolution remote sensing images, and perform preprocessing to lay the data foundation for subsequent analysis; preprocessing includes cleaning, format conversion, and spatial interpolation of non-image data (such as meteorological and soil data) to eliminate data noise, unify data format, and improve data spatial resolution; and performing radiometric correction, geometric correction, and image enhancement on remote sensing image data to eliminate image distortion, improve image contrast, and enhance interpretability.

[0025] S2. Extreme climate events are introduced as an independent evaluation dimension. By establishing a correlation mechanism model between extreme climate events and gully erosion, the inducing and aggravating effects of extreme climate on gully erosion are quantitatively analyzed.

[0026] S3. Construct an indicator system for assessing the intensity of anthropogenic disturbance to quantify the impact of human activities on ridge erosion. The assessment of anthropogenic disturbance intensity specifically quantifies the impact of human activities on ridge erosion, clarifying the intensity and mechanism of anthropogenic disturbance. First, an indicator system for assessing the intensity of anthropogenic disturbance is constructed, specifically including land use change intensity, engineering construction intensity, and vegetation destruction intensity, to characterize the intensity of regional anthropogenic disturbance. Second, multi-criteria decision-making methods such as the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation are used to quantitatively assess the intensity of anthropogenic disturbance in the ridge erosion area. Finally, methods such as geographic detectors and structural equation modeling are used to analyze the contribution and action path of anthropogenic disturbance intensity to ridge erosion, clarifying the impact mechanism of anthropogenic disturbance intensity on ridge erosion.

[0027] S4. Extract landslide features from the preprocessed remote sensing images, and verify the landslide features in the field to ensure the accuracy of the feature data; at the same time, conduct detailed investigations in conjunction with the micro-geomorphic features of landslides such as catchment slopes, landslide walls, alluvial cones, gullies, and alluvial fans to supplement and improve the landslide feature information.

[0028] S5. Based on three dimensions—extreme climate events, intensity of anthropogenic disturbance, and characteristics of ridge collapse—a comprehensive evaluation index system for the hazard of ridge collapse erosion is constructed, which includes an extreme climate impact index, anthropogenic disturbance intensity index, ridge collapse morphology index, and ridge collapse erosion sediment yield index, thereby achieving a comprehensive integration of internal and external factors.

[0029] S6. Construct a deep learning model that integrates a Convolutional Neural Network (CNN) module, a Recurrent Neural Network (RNN) module, and a Transformer module with a self-attention mechanism. Based on the aforementioned ridge erosion hazard assessment index system and the deep learning model, comprehensively evaluate the ridge erosion hazard, obtain the ridge erosion hazard assessment results, and classify the hazard levels.

[0030] S7. Based on the assessment results of the hazards of gully erosion, and combined with factors such as the leading industries of the township and the economic feasibility of governance, a urgency assessment model for gully erosion governance is constructed to obtain the urgency assessment results, classify the urgency levels, and formulate governance priorities.

[0031] S8. Using Geographic Information System (GIS) technology, the results of the assessment of the hazards and urgency of gully erosion are spatially visualized to generate thematic distribution maps, which intuitively present the key areas for gully erosion control. Based on the visualization results and assessment analysis conclusions, targeted gully erosion prevention and control measures and policy recommendations are proposed to provide a scientific basis for regional gully erosion control decisions.

[0032] S9. Based on the assessment results of the erosion hazards of the collapsed hills and the spatial interpolation data, a multi-objective optimization model is constructed with the optimization objectives of governance benefits and governance costs, and the parameters of the hill collapse governance measures are optimized.

[0033] S10. Construct economic benefit models for hill collapse management under different leading industry types, and assess the economic feasibility of hill collapse management.

[0034] This solution integrates extreme climate drivers, deep learning evaluation, and the feasibility of township industries and economies into a unified system. It introduces extreme climate as an independent dimension in hilly area assessment, addressing the problem of traditional assessments neglecting external driving factors and leading to biased results. By constructing a deep learning model integrating CNN, RNN, and Transformer, it achieves comprehensive extraction of spatial, temporal, and global correlation features of hilly areas, improving evaluation accuracy. Furthermore, it incorporates the township's leading industries and the economic feasibility of governance into the urgency model and constructs industry-specific economic benefit models, ensuring that governance decisions consider both ecological and economic benefits. This systematically solves the problems of one-sidedness in existing hilly area assessments, unclear governance priorities, and a lack of targeted measures, achieving a scientific, precise, and implementable process throughout.

[0035] In this embodiment, the data types and parameters of the acquired data are specifically defined. Meteorological data includes rainfall, rainfall intensity, temperature, wind speed, etc., providing basic parameters for the identification of extreme climate events; topographic data includes elevation, slope, aspect, etc., used to analyze the topographic conditions for ridge development; soil data includes soil type, soil texture, soil organic matter content, etc., used to assess soil erodibility; vegetation data includes vegetation type, vegetation coverage, etc., used to analyze the soil-fixing and water-retaining effects of vegetation; human disturbance data includes land use type, engineering construction activities, etc., used to quantify the intensity of human activities; remote sensing imagery includes high-resolution satellite imagery with a spatial resolution of not less than 0.5 meters and UAV aerial imagery, ensuring the accuracy of ridge boundary identification and micro-topography extraction.

[0036] The data types in this scheme cover both natural environmental factors and human activity factors that influence the development of gully collapse, forming the basic dataset required for gully collapse assessment. Through refined multi-source data definition, the input basis for subsequent assessment is ensured to be comprehensive and reliable, thereby improving the overall assessment accuracy and stability.

[0037] In this embodiment, the method for constructing the correlation mechanism model between extreme climate events and ridge erosion is further defined. First, based on preprocessed meteorological data, extreme value theory and frequency analysis methods are used to identify extreme climate events such as rainstorms and typhoons that affect ridge erosion, and their recurrence periods and occurrence patterns are statistically analyzed. Second, using the extreme climate events as input variables, a joint probability distribution model of extreme climate events and ridge erosion amount is established using a Copula function. The Markov chain method is used to simulate the state transition probabilities of different extreme climate events, simulating the dynamic impact process of extreme climate on ridge erosion. Finally, the joint probability distribution model is used to analyze the impact mechanism of extreme climate events on the ridge erosion process, quantifying the contribution of the frequency, intensity, and duration of extreme climate events to the amount of ridge erosion, thus elevating the impact of extreme climate from a qualitative description to a quantitative expression.

[0038] This scheme probabilistically and quantitatively characterizes the induction mechanism of extreme climate and ridge erosion. Extreme value theory and frequency analysis solve the problems of identifying extreme climate events and calculating their return periods. Copula functions can reflect the nonlinear and asymmetric correlation between extreme rainfall and ridge erosion, which is more consistent with actual physical processes than traditional linear correlation analysis. Markov chains can simulate the state transition probability of extreme climate events, reflecting the cumulative and hysteresis effects of extreme climate on ridge erosion. By combining these two techniques, the induction and aggravation mechanisms of extreme climate on ridge erosion can be quantitatively expressed, filling the gap in the qualitative description of extreme climate factors in traditional ridge erosion assessment.

[0039] In this embodiment, the specific components of the gully collapse characteristics are further defined, and specific indicators of morphology, erosion and sediment yield, and development stage are clarified. Morphological characteristics include area, depth, volume, slope, perimeter, and aspect, which represent the scale of the gully collapse from the dimensions of geometric size and spatial form; erosion and sediment yield characteristics include erosion modulus and sediment yield, which quantify the degree of soil and water loss from the dimension of erosion intensity; development stage characteristics include active period and stable period, which determine the development trend of the gully collapse from the dimension of evolution stage.

[0040] Specifically, a combination of visual interpretation and automatic computer interpretation is used to interpret the pre-processed remote sensing images and accurately identify the boundaries and extent of the landslides. Remote sensing image processing software such as ENVI and ArcGIS are used to extract the morphological features (area, depth, volume, slope, perimeter, etc.), erosion and sediment yield characteristics (erosion modulus, sediment yield, etc.), and development stage characteristics (active period, stable period, etc.) of the landslides.

[0041] This scheme, by comprehensively extracting multi-dimensional features of gully erosion, can more accurately characterize the current status and development level of gully erosion. Morphological features such as area and volume are intuitive manifestations of the scour's harmfulness and directly affect the scale and cost of control engineering measures. Erosion modulus and sediment yield are quantitative indicators of erosion intensity, reflecting the activity level of the gully erosion and its contribution to downstream sediment. The division of active / stable periods helps to determine the urgency of control timing and the pertinence of control measures. It also provides rich basic data for the subsequent construction of an evaluation index system.

[0042] In this embodiment, the construction method of the evaluation index system for the hazard of hill collapse erosion is further defined. The weights of each evaluation index are determined by combining the analytic hierarchy process (AHP) and the entropy weight method: the AHP incorporates the experience and judgment of domain experts, calculating subjective weights by constructing a judgment matrix; the entropy weight method calculates objective weights based on the dispersion of each index's data, with greater dispersion indicating a larger amount of information contained in the index and thus a higher weight; the combination of the two achieves a comprehensive assessment of subjective and objective weights. Simultaneously, grey relational analysis and principal component analysis are used to optimize the index system: grey relational analysis calculates the correlation between each index and the hazard of hill collapse, eliminating redundant indicators with low correlation; principal component analysis extracts principal components with a cumulative variance contribution rate greater than a preset threshold, and based on the loading matrix of the principal components, the retained evaluation indicators are clustered and reorganized to obtain an optimized evaluation index system.

[0043] This scheme objectively combines weighting and indicator optimization to reduce subjective bias, eliminate redundant information, and improve the scientific rigor and stability of the evaluation system. The combined weighting method of the analytic hierarchy process (AHP) and entropy weighting takes into account both the prior knowledge of domain experts and the inherent informational characteristics of the data, avoiding the limitations of a single weighting method. Grey relational analysis eliminates redundant indicators with low correlation to the severity of the collapse, reducing information overlap between indicators. Principal component analysis extracts core evaluation dimensions, reducing the complexity of the indicator system and improving evaluation efficiency.

[0044] This embodiment defines the construction and verification method of a deep learning model integrating a CNN module, an RNN module, and a self-attention mechanism neural network module. The CNN module extracts the spatial distribution and morphological features of the landslide ridges, such as ridge boundaries, area, and slope, through convolutional and pooling layers. The RNN module uses a Long Short-Term Memory (LSTM) network structure to extract the temporal variation features of extreme weather events and human disturbances, capturing long-term dependencies in time-series data. The self-attention mechanism-based Transformer module extracts the global correlation features between various evaluation indicators, models the complex nonlinear relationships between multiple indicators, performs adaptive weighted fusion through a fusion layer, and outputs the landslide erosion hazard assessment result after inputting into a fully connected layer. During model training, ensemble learning and transfer learning methods are combined to improve prediction accuracy and generalization ability, and sensitivity analysis and uncertainty analysis methods are used to verify the accuracy and reliability of the evaluation results.

[0045] Specifically, a large amount of data related to gully erosion was collected, including extreme climate data, anthropogenic disturbance data, gully erosion characteristic data, and existing hazard assessment results. These data were proportionally divided into training and testing sets. The backpropagation algorithm was used to train the model, adjusting its parameters and structure. Ensemble learning and transfer learning methods were combined to improve the model's prediction accuracy and generalization ability. For the Transformer model, specific parameter optimization operations were performed sequentially, including data preprocessing, model initialization, parameter training, optimization algorithm adjustment, validation set evaluation, and production environment deployment. Optimization algorithms included Stochastic Gradient Descent (SGD), Adaptive Moment Estimation (Adam), and Root Mean Square Propagation (RMSProp). Finally, the optimized evaluation index system data was input into the trained deep learning model to complete the quantitative comprehensive evaluation of the gully erosion hazard, obtaining the gully erosion hazard assessment results.

[0046] Based on the assessment results of the hazard of ridge erosion, the hazard of ridge erosion is divided into four levels: mild, moderate, severe, and extremely severe, so as to achieve a direct distinction of the hazard of ridge erosion. Sensitivity analysis and uncertainty analysis are used to verify and analyze the assessment results, evaluate the impact of changes in indicators on the assessment results, and improve the accuracy and reliability of the assessment results.

[0047] In this scheme, the CNN module effectively extracts the spatial morphological features of eroded hills, solving the problem of traditional methods struggling to quantify the geometric characteristics of eroded hills; the RNN module captures the temporal evolution patterns of extreme climate and human disturbances, reflecting the dynamic changes in erosion; and the Transformer module models the complex nonlinear relationships between multiple indicators, overcoming the limitations of traditional linear weighted models. The weighted fusion of these three modules achieves a multi-dimensional and multi-level comprehensive evaluation of the hazards of eroded hills, with significantly higher evaluation accuracy than a single model. The introduction of ensemble learning and transfer learning solves the model training problem under limited eroded hill sample conditions, enabling the model to maintain good generalization ability even with small sample sizes; sensitivity analysis and uncertainty analysis ensure the reliability and stability of the evaluation results.

[0048] In this embodiment, an assessment model for the urgency of gully erosion control is constructed based on factors such as the assessment results of the hazard assessment of gully erosion, the frequency of extreme weather events, the intensity of human disturbance, regional ecological sensitivity, regional socio-economic importance, township leading industries, and the economic feasibility of control. Specifically, the hazard assessment results of gully erosion reflect the severity of the gully erosion itself; the frequency of extreme weather events reflects the intensity of future external triggers; the intensity of human disturbance reflects the sustained impact of human activities; regional ecological sensitivity reflects the ecological protection value of control; regional socio-economic importance reflects the socio-economic significance of control; township leading industries reflect the correlation between control and regional economic development; and the economic feasibility of control reflects the financial input and output benefits of control.

[0049] A judgment matrix is ​​constructed, and the subjective weights of each evaluation factor are calculated using the analytic hierarchy process (AHP). A fuzzy comprehensive evaluation method is used, based on the subjective weights and combined with the measured data of each evaluation factor, to calculate the objective weights of each evaluation factor. Based on the subjective and objective weights, the combined weights of the evaluation factors are comprehensively determined. The urgency of addressing the collapse of work sites is quantitatively assessed using a collapse site management urgency assessment model, and the urgency is divided into four levels: low, medium, high, and extremely high. Based on the results of these urgency levels, a priority plan for addressing the collapse of work sites is formulated.

[0050] Specifically, the weighting of the assessment factors is as follows: erosion severity of gully collapse 0.3, frequency of extreme weather events 0.2, intensity of human disturbance 0.2, ecological sensitivity 0.1, socio-economic importance 0.1, township's leading industry 0.07, and economic feasibility of remediation 0.03. A priority remediation plan for gully collapses is formulated, prioritizing those with higher urgency. Differentiated remediation approaches are proposed for gullies with different urgency levels; for example, gullies with extremely high urgency will be remediated using a combination of engineering and biological methods, while gullies with low urgency will be remediated primarily through natural restoration.

[0051] This plan incorporates the township's leading industries and the economic feasibility of governance into the urgency assessment model for addressing hill erosion, aligning governance decisions with the needs of regional economic development. The weighting of seven factors reflects a scientific understanding of the influencing factors in hill erosion governance decisions: the severity of hill erosion serves as the core foundation, determining the necessity of governance; extreme weather and human disturbance, as external driving factors, reflect the urgency of governance; ecological sensitivity and socio-economic importance, as protection needs, embody the comprehensive value of governance; and the township's leading industries and the economic feasibility of governance, as development guidelines, ensure the sustainability of governance. The division into four urgency levels allows governance resources to be allocated to the areas most in need, avoiding resource waste; the formulation of priority plans provides management departments with clear decision-making basis, realizing a shift from "passive governance" to "proactive prevention and control."

[0052] In this embodiment, a parameter optimization method for the governance measures is defined, covering the overall optimization of hydraulic structures, vegetation configuration, and genetic algorithms. Based on spatial interpolation data, the value ranges of decision variables for the engineering and biological measure parameter sets are determined. A multi-objective fitness function is constructed with the optimization objectives of maximizing governance benefits and minimizing governance costs. A genetic algorithm is used to encode the decision variables, and the Pareto optimal solution set is obtained by iteratively solving the multi-objective fitness function. The optimal parameter combination for the gully collapse governance measures is determined based on the Pareto optimal solution set.

[0053] Specifically, based on spatial interpolation data (such as the spatial distribution of topographic slope, catchment area, and soil permeability), the location, size, and slope parameters of intercepting ditches, retaining walls, and drainage works are optimized to ensure that intercepting ditches can intercept all slope runoff, retaining walls can effectively prevent landslides, and drainage works can promptly remove accumulated water from the area. Simultaneously, the selection of tree species, planting density, and planting time parameters for vegetation restoration are optimized. Native and suitable tree species are prioritized for the landslide-affected area, and planting density and time are determined based on local climate conditions and soil characteristics. On this basis, multi-objective optimization algorithms and genetic algorithms are used to comprehensively optimize the engineering and biological measures for landslide control, taking into account control effectiveness, engineering costs, and ecological benefits, thereby improving the overall effectiveness and efficiency of landslide control.

[0054] This scheme introduces multi-objective optimization algorithms and genetic algorithms into the optimization of gully erosion control engineering measures, realizing the transformation from empirical design of single measures to systematic optimization of multiple measures. Spatial interpolation data ensures that the location and size parameters of engineering measures can adapt to local topographic and hydrological conditions; the selection of native and suitable tree species guarantees the survival rate and ecological adaptability of vegetation restoration, avoiding the risk of invasion by alien species; the multi-objective optimization algorithm takes into account three objective functions: control effect, engineering cost, and ecological benefits, while the genetic algorithm searches for the global optimum through operations such as selection, crossover, and mutation.

[0055] In this embodiment, the construction method of the industry-specific economic benefit model is defined as follows: First, based on the urgency assessment results of the hill collapse management and the optimized parameter combination of the hill collapse management measures, the type and scale of the hill collapse management engineering measures are determined; industrial structure data of the township where the hill collapse occurred are obtained to determine the dominant industry type; second, the optimized parameter combination and the dominant industry type are spatially overlaid to determine the potential impact range of the management measures on different industries; the total management cost is calculated according to the determined engineering measure type and scale; the total management benefit is predicted according to the potential impact range and industry output data; finally, dynamic economic evaluation indicators are calculated based on the total management cost and total management benefit; the dynamic economic evaluation indicators include net present value, internal rate of return, and investment payback period; the dynamic economic evaluation indicators are fed back to the multi-objective optimization model as constraints for parameter optimization.

[0056] Specifically, the dominant industry type includes at least one of agriculture, forestry, animal husbandry, fishery, industry, and tourism; the total cost of governance includes one-time engineering construction costs, vegetation restoration costs, and annual monitoring and maintenance costs; the total benefits of governance include agricultural production increase benefits for agriculture-dominated townships, forestry income increase benefits for forestry-dominated townships, animal husbandry production increase benefits for animal husbandry-dominated townships, fishery production increase benefits for fishery-dominated townships, industrial emission reduction benefits for industry-dominated townships, and tourism income increase benefits for tourism-dominated townships.

[0057] This scheme constructs an economic benefit model for hillside remediation based on industry type, combining ecological governance with industrial development, and realizing a shift from "pure ecological governance" to "ecological and economic synergy governance." For different leading industry types, corresponding benefit calculation methods are set, making the economic benefit assessment more realistic: for agricultural-led townships, the focus is on calculating the increased yield of cash crops, reflecting the promoting effect of governance on agricultural production; for tourism-led townships, the focus is on calculating tourism revenue brought by increased visitor traffic, reflecting the driving effect of governance on the tourism industry; for industrial-led townships, the focus is on calculating emission reduction benefits, reflecting the supporting effect of governance on sustainable industrial development. Net present value (NPV), internal rate of return (IRR), and payback period are used to assess the economic feasibility of the remediation project from different dimensions: NPV reflects the absolute return level of the project, IRR reflects the relative return efficiency of the project, and payback period reflects the speed of capital recovery. The combination of these three indicators provides a comprehensive quantitative basis for decision-making regarding remediation funding.

[0058] In this embodiment, the economic benefit model is extended by adding an analysis of the differences in its quantitative application across different townships. First, natural geographical data and socioeconomic statistics of the townships where the landslides occurred are collected to construct township feature vectors. Second, these township feature vectors are input into the multi-objective optimization model. By adjusting the cost and benefit coefficients in the multi-objective optimization model, the landslide management costs and benefits for different townships are calculated. Third, a comparison matrix between townships is established, including net present value, internal rate of return, and investment payback period. Based on the differences in the comparison matrix, key influencing factors affecting economic feasibility are identified. Finally, based on these key influencing factors, the weights of the human disturbance intensity assessment index system are dynamically adjusted, or the optimal parameter combination of the landslide management measures is adjusted.

[0059] Specifically, natural geographical data and socio-economic statistics include topography, climate conditions, industrial structure, economic development level, and fiscal revenue.

[0060] This plan enhances the practicality and scalability of its approach through refined governance and differentiated policies, providing a scientific basis for developing personalized governance plans for different townships. It considers the impact of varying natural conditions and economic development levels across townships on the economic benefits of governance, achieving site-specific optimization of the plan. By establishing a comparison matrix between townships, key factors affecting the economic feasibility of governance can be identified, allowing for targeted adjustments to the plan to maximize economic benefits.

[0061] The present invention also provides the following specific embodiments: Meizhou City, located in northeastern Guangdong Province, is one of the areas in Guangdong most severely affected by gully erosion, with a total gully erosion area of ​​435.5 km². 2 This area accounts for more than 50% of the gully erosion area in Guangdong Province. The soil in the region is mainly red soil, and the deep granite weathering layer is prone to forming gully erosion landforms under high temperature and continuous rainfall conditions, making it suitable as the implementation area of ​​this invention.

[0062] 1. Data Collection and Preprocessing 1.1 Scope of Data Collection: The data covers the entire Meizhou City area, spanning from 2000 to 2025 (the last 25 years), ensuring the timeliness and continuity of the data.

[0063] 1.2 Specific data collection: Meteorological data: collected from the China Meteorological Data Network and Meizhou Meteorological Bureau, including annual rainfall, monthly rainfall, daily maximum rainfall, rainfall intensity (1h, 6h, 24h), annual average temperature, extreme high temperature, extreme low temperature, wind speed, typhoon path and impact intensity, etc., in Excel and TXT formats.

[0064] Topographic data: 1:10000 scale digital elevation model (DEM) data with a spatial resolution of 30m was downloaded from the geospatial data cloud platform of the Ministry of Natural Resources, and aerial survey data from UAVs was used to extract the micro-topographic features of the collapsed hills.

[0065] Soil data: Collection of the second soil survey report of Meizhou City and soil sampling data (sampling density of 10 km²). 2 One sampling point was used, including soil type, soil texture (sandy, loamy, clay), soil organic matter content, soil bulk density, soil erosion resistance coefficient, etc. The data format was Shp and Excel.

[0066] Vegetation data: Sentinel-2 L1C level multispectral vegetation data (spatial resolution 10m) and Moderate-resolution Imaging Spectroradiometer (MODIS) data (spatial resolution 250m) were collected from 2016 to 2025 to calculate vegetation cover. At the same time, the vegetation types in the core implementation area were investigated in the field, and the species and distribution of trees, shrubs and herbaceous plants were recorded.

[0067] Human disturbance data: Land use type data (Shp format, updated every 5 years), project filing data (including highway construction, mining, real estate development, etc.), vegetation destruction penalty records, and deforestation and land reclamation survey data are collected from the Meizhou Municipal Bureau of Natural Resources, Bureau of Agriculture and Rural Affairs, and Bureau of Housing and Urban-Rural Development.

[0068] Remote sensing imagery: GF-02 satellite imagery (spatial resolution 0.8m) and WorldView-3 satellite imagery (spatial resolution 0.3m) were collected. At the same time, a DJI Phantom 4 RTK drone was used to conduct aerial photography of the area, obtaining drone aerial images with a spatial resolution of 0.1m. The aerial photography was conducted in October and November each year (when vegetation cover is relatively stable and there is no rainstorm interference), at an altitude of 150m, with a forward overlap of 80% and a lateral overlap of 70%.

[0069] 1.3 Data Preprocessing Operations: Non-image data: Data cleaning was performed using Excel and SPSS software to remove missing and outlier values ​​(using 3D printing). (Based on the principle of elimination); ArcGIS 10.8 software was used to convert Excel and TXT format data to Shp and Grid formats; Kriging interpolation was used to perform spatial interpolation to convert discrete data such as meteorological and soil data into spatially continuous data, with an interpolation resolution of 30m to ensure spatial consistency of the data.

[0070] Remote sensing image preprocessing: ENVI 5.6 software was used for processing. First, radiometric correction (using the FLAASH atmospheric correction method) was performed to eliminate the influence of atmospheric scattering and absorption on the image. Then, geometric correction was performed (using the quadratic polynomial correction method, with control points selected from obvious features such as road intersections and field bends, with no fewer than 20 control points, and the correction error controlled within 1 pixel). Finally, image enhancement was performed (using histogram equalization and Gaussian filtering methods) to improve image contrast and clarity, eliminate image noise, and facilitate the identification of landslide boundaries and feature extraction.

[0071] 2. Identification and Analysis of Extreme Climate Events 2.1 Criteria for identifying extreme weather events: Based on the climate characteristics of Meizhou City, extreme rainstorm events are defined as daily rainfall ≥100mm or 24-hour rainfall ≥150mm, and extreme typhoon events are defined as typhoons affecting Meizhou City with a typhoon level ≥8 (wind speed ≥17.2m / s).

[0072] 2.2 Implementation of the identification method: The POT (Threshold Excess) model from extreme value theory was used, with a threshold of 100 mm of daily rainfall, to identify extreme rainstorm events. Frequency analysis (Pearson Type III distribution) was used to calculate the recurrence intervals (5 years, 10 years, 20 years, and 50 years) of extreme climate events.

[0073] 2.3 Feature Analysis: The frequency (annual average number of occurrences), intensity (average rainfall, average wind speed), and duration (number of days of extreme rainstorms, number of days of typhoon impact) of extreme rainstorms and extreme typhoons in Meizhou City from 2000 to 2025 were statistically analyzed. Statistical charts were generated using Excel and Origin software to analyze their temporal trends and spatial distribution characteristics.

[0074] 2.4 Analysis of the Influence Mechanism: A joint probability distribution model of extreme torrential rainfall and ridge erosion was established using the Copula function (Gumbel-Copula function selected). The Markov chain method was used to analyze the state transition probability of extreme climate events (e.g., the probability of active ridge erosion state transition after torrential rainfall). Correlation analysis was performed using SPSS software to quantify the correlation coefficient between extreme climate events and ridge erosion (R² required). 2(≥0.6), clarifying the induction and aggravation mechanism of extreme climate on gully erosion in Meizhou City.

[0075] 3. Assessment of the intensity of human disturbance 3.1 Construction of the indicator system: Construct an indicator system for assessing the intensity of anthropogenic disturbance, specifically including the intensity of land use change, the intensity of engineering construction, and the intensity of vegetation destruction: Land use change intensity: calculated using land use dynamics, the formula is as follows: ,in , These refer to the land use area at the beginning and end of the research period, respectively. The research period is defined as the number of years.

[0076] Construction intensity: It is calculated based on the proportion of construction area, and the formula is C = construction area / total area of ​​the region × 100%. The construction area is obtained from remote sensing image interpretation and project filing data.

[0077] Vegetation destruction intensity: Two indicators were used: vegetation coverage change rate and the proportion of vegetation destruction area. Vegetation coverage change rate = (NDVI at the end of the period - NDVI at the beginning of the period) / NDVI at the beginning of the period × 100%.

[0078] 3.2 Strength Assessment: The Analytic Hierarchy Process (AHP) was used to determine the weights of the indicators. Five senior experts in the field of soil and water conservation were invited to score the indicators and construct a judgment matrix. The consistency test CR < 0.1. The fuzzy comprehensive evaluation method was used to classify the intensity of human disturbance into five levels: weak, relatively weak, medium, relatively strong, and strong. The evaluation matrix was constructed using triangular fuzzy numbers, and the evaluation results were calculated using MATLAB software.

[0079] 3.3 Analysis of the Influence Mechanism: Using geographic detector software, the q-statistics (q∈[0,1]) of each anthropogenic disturbance factor (land use change, engineering construction, vegetation destruction) were calculated. The larger the q value, the greater the impact of the factor on ridge erosion (q≥0.3 required). A structural equation model was constructed using AMOS software to analyze the interaction between the anthropogenic disturbance factors and their effects on ridge erosion. The model fit indexes were: Root Mean Square Error of Approximation (RMSEA) <0.08 and Comparative Fit Index (CFI) >0.9, clarifying the impact mechanism of deforestation and other behaviors on ridge erosion.

[0080] 4. Feature extraction of the collapse itself 4.1 Image Interpretation: A combination of visual interpretation and automatic computer interpretation was adopted, based on preprocessed satellite imagery and UAV aerial imagery, and the interpretation was performed in ArcGIS 10.8 software.

[0081] Automatic computer interpretation: The Support Vector Machine (SVM) classification method was used to construct a sample library for interpreting landslides (with no less than 500 samples covering different types and developmental stages of landslides). The interpretation accuracy was verified by a Kappa coefficient ≥ 0.85.

[0082] Visual interpretation: Corrects the automatic interpretation results, focusing on correcting blurred boundaries and small-scale hill collapses (area <100m²). 2 The interpretation results ensure that the accuracy of the collapse boundary identification is ≥90%.

[0083] 4.2 Feature Extraction: The features of the landslide were extracted using ENVI 5.6 and ArcGIS 10.8 software, as detailed below: Morphological characteristics: area (calculated using ArcGIS spatial analysis tools), depth (calculated using DEM data, collapse depth = collapse top elevation - collapse bottom elevation), volume (calculated using the irregular triangular prism method), slope (calculated using the Digital Elevation Model (DEM) slope extraction tool, slope grades are 0~15°, 15~30°, 30~45°, >45°), perimeter, and aspect.

[0084] Erosion and sediment yield characteristics: erosion modulus (calculated using the USLE soil loss equation, formula A=R×K×LS×C×P, where R is the rainfall erosivity factor, K is the soil erodibility factor, LS is the topographic factor, C is the vegetation cover factor, and P is the soil and water conservation measures factor), and sediment yield (erosion modulus × erosion area).

[0085] Developmental stage characteristics: Based on the morphology and erosion intensity of the landslide, it is divided into an active period (steep landslide, strong erosion, no obvious vegetation cover) and a stable period (gentle landslide, weak erosion, vegetation cover ≥60%), which is verified by field investigation.

[0086] 4.3 Field verification: Fifty typical landslide ridges (covering different sizes and development stages) in the core implementation area were selected for field investigation. The area, depth, slope and other characteristics of the landslide ridges were measured using a total station and compared with the image extraction results, with the error controlled within 5%. At the same time, micro-geomorphic features such as the slope of the catchment slope, the height of the landslide wall, the thickness of the landslide body, the width of the gully, and the area of ​​the alluvial fan were recorded to supplement and improve the landslide ridge feature database.

[0087] 5. Construction of Evaluation Index System 5.1 System Construction: Based on three dimensions—extreme climate events, intensity of human disturbance, and characteristics of gully erosion—a four-level evaluation index system is constructed. The first-level index is the gully erosion hazard assessment index; the second-level indexes are the extreme climate impact index, the intensity of human disturbance index, the gully erosion morphology index, and the gully erosion sediment yield index; the third- and fourth-level indicators are set in conjunction with the regional characteristics of Meizhou City, and the quantitative standards for each indicator are clearly defined.

[0088] 5.2 Weight Determination: The analytic hierarchy process (AHP) (subjective weighting) and the entropy weighting method (objective weighting) are combined, with subjective weights accounting for 40% and objective weights accounting for 60%.

[0089] Analytic Hierarchy Process (AHP): Five experts in soil and water conservation and ecological environment were invited to score the data, construct a judgment matrix, and calculate the subjective weights based on the consistency ratio (CR) of <0.1.

[0090] Entropy weight method: SPSS software is used to standardize the data of each indicator (range standardization), calculate the entropy value and entropy weight of each indicator, and obtain the objective weight.

[0091] Overall weight: Overall weight = subjective weight × 40% + objective weight × 60%, with a focus on key factors such as geological type (weathered granite layer) and slope (>30°), and its overall weight is not less than 0.1.

[0092] 5.3 System Optimization: Grey relational analysis was used to calculate the correlation between each indicator and the hazard of gully erosion, and redundant indicators with a correlation of <0.5 were removed. Principal component analysis was used to reduce the dimensionality of the remaining indicators and extract principal components (cumulative variance contribution rate ≥85%). The optimized evaluation indicator system includes 12 fourth-level indicators to ensure the scientificity and effectiveness of the system.

[0093] 6. Deep learning model construction and training for comprehensive evaluation of the hazards of gully erosion. 6.1 Model Construction: Based on the open-source frameworks TensorFlow and PyTorch, a deep learning evaluation model fusion of CNN, RNN, and Transformer is constructed. The model structure is as follows: CNN module: Adopting the AlexNet architecture, it contains 5 convolutional layers and 3 pooling layers, used to extract the spatial features (morphology and spatial distribution) of the collapsed hills. RNN module: Adopting LSTM architecture, containing 3 hidden layers, used to extract temporal features (time change trends) of extreme weather and human disturbances. Transformer module: contains 6 encoder layers, 6 decoder layers, and 8 attention heads, used to extract global correlation features of various metrics; Fusion layer: The weighted summation method is used to fuse the output features of the three modules, input them into the fully connected layer, and output the assessment results of the hazard of gully collapse erosion.

[0094] 6.2 Dataset Construction: Data on 2,300 landslides in Meizhou City (covering different levels of damage and different areas) were collected and divided into a training set (1,610 data points), a test set (460 data points), and a validation set (230 data points) in a ratio of 7:2:1. The data were standardized (Z-score standardization) and the SMOTE algorithm was used to solve the problem of imbalanced datasets (by expanding a small number of extreme and severe landslide samples).

[0095] 6.3 Model Training and Optimization: Training parameters: initial learning rate of 0.001, adjusted using cosine annealing; batch size of 32; number of epochs of 100; cross-entropy loss function.

[0096] Model optimization: Combining ensemble learning (random forest ensemble) and transfer learning (pre-trained models using the ImageNet dataset) to improve model prediction accuracy and generalization ability.

[0097] Transformer model optimization: Data preprocessing (normalization, standardization), model initialization (random weight initialization), parameter training (backpropagation algorithm), optimization algorithm adjustment (using Adam algorithm), validation set evaluation (calculating accuracy, recall, and F1 score), and production environment deployment (converting to ONNX format for easy subsequent application).

[0098] Model validation: The accuracy of the test set is ≥90%, and the accuracy of the validation set is ≥88%, ensuring that the model's prediction accuracy meets the needs of practical applications.

[0099] 6.4 Overall Evaluation: The optimized evaluation index system data (after standardization) is input into the trained deep learning model to calculate the hazard assessment scores (out of 100) of 2,300 collapsed hills in Meizhou City in batches.

[0100] 6.5 Level Classification: Using the natural breakpoint method, the evaluation scores are divided into four levels, with specific criteria as follows: Mild hazard: Score 0-25, landslide area <500m² 2 erosion modulus <5000t / (km)2 a) Vegetation coverage ≥ 60%; Moderate hazard: Score 25-50, landslide area 500-2000m² 2 Erosion modulus 5000–15000 t / (km²) 2 a) Vegetation coverage of 30%–60%; Severe hazard: Score 50-75, landslide area 2000-5000m² 2 Erosion modulus 15000~30000 t / (km) 2 a) Vegetation coverage of 10%–30%; Extremely severe hazard: Score 75-100, landslide area > 5000m² 2 Erosion modulus > 30000 t / (km) 2 •a) Vegetation coverage <10%.

[0101] 6.6 Result Verification: Sensitivity analysis was used to apply sensitivity to each input feature sequentially. The perturbation was observed, and the rate of change of the evaluation results was observed. The rate of change was <10%, indicating that the model has good stability. Uncertainty analysis was adopted, and the Monte Carlo simulation method was used (1000 simulations) to calculate the coefficient of variation (CV<0.1) of the evaluation results to ensure the accuracy and reliability of the evaluation results.

[0102] 7. Analysis of the Urgency of Governance 7.1 Construction of the governance urgency assessment model: Based on the hazard assessment results of gully erosion in Meizhou City, and considering seven factors including the frequency of extreme weather events in the region, the intensity of human disturbance, ecological sensitivity, socio-economic importance, the leading industries of townships, and the economic feasibility of remediation, a linear weighted summation model is constructed. The model calculation formula is as follows: ,in To score the urgency of governance, For the first Each factor weight, For the first Standardized scores for each factor (0-100 points).

[0103] Specific quantitative standards for each factor: Harmfulness of gully erosion: The aforementioned evaluation scores are directly standardized to 0-100 points.

[0104] Extreme weather frequency: The average number of extreme weather events per year, scored after standardization; the more frequent the events, the higher the score.

[0105] Human disturbance intensity: The aforementioned human disturbance intensity evaluation score is directly standardized.

[0106] Ecological sensitivity: Based on the distribution of ecological protection red lines, water source protection areas and nature reserves in Meizhou City, it is divided into extremely high sensitivity, high sensitivity, medium sensitivity, low sensitivity and extremely low sensitivity, with corresponding scores of 100, 80, 60, 40 and 20.

[0107] Socioeconomic importance: based on population density (people / km²) 2 The distance to major transportation routes (km) and the distance to towns (km) are quantified. The higher the population density and the closer the distance, the higher the score.

[0108] Townships with leading industries: those with agriculture, forestry, and tourism as their leading industries score higher (70-100 points); those with industry as their leading industry score moderately (50-70 points).

[0109] Economic feasibility of governance: The cost-benefit ratio (benefit / cost) is used for quantification. The higher the rate of return, the higher the score (0-100 points).

[0110] 7.2 Weight Determination: Using a combination of the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation, five experts in soil and water conservation and engineering management were invited to score the factors and determine their weights. The final weight allocation was as follows: 0.3 for the hazard of gully erosion, 0.2 for the frequency of extreme weather events, 0.2 for the intensity of human disturbance, 0.1 for ecological sensitivity, 0.1 for socio-economic importance, 0.07 for the leading industries of the township, and 0.03 for the economic feasibility of the management.

[0111] 7.3 Urgency Assessment and Solution Development 7.3.1 Urgency Assessment: Using the above model, the urgency score (0-100 points) for the treatment of 2,300 collapsed hills in Meizhou City was calculated, and the natural breakpoint method was used to classify them into four levels: (1) Low urgency: Score 0-25 points, mainly stable period, mild hazard of hill collapse, distributed in areas with low ecological sensitivity and sparse population; (2) Medium urgency: Score 25-50 points, mainly moderate hazard of hill collapse, distributed in areas with moderate ecological sensitivity and weak human disturbance; (3) High urgency: Score 50-75 points, mainly severe erosion, distributed in areas with high ecological sensitivity and strong human disturbance; (4) Extremely high urgency: 75-100 points, mainly extremely serious landslides, distributed in ecologically sensitive areas (water sources, nature reserves), densely populated areas, and near main transportation routes, and where extreme weather occurs frequently and human disturbance is strong.

[0112] 7.3.2 Governance Priority Scheme The overall governance order is determined by "extremely high urgency > high urgency > medium urgency > low urgency". Within the same level, the order is further sorted by "ecological sensitivity - socio-economic importance - dominant industry type". Priority is given to the governance of hill erosion in water source areas, nature reserves, densely populated areas and major agricultural production areas, so as to achieve precise classification and priority governance of hill erosion.

[0113] 8. Visualization of evaluation results Using GIS technology, the distribution of hazard levels and urgency levels of gully erosion are spatially visualized, generating thematic distribution maps that intuitively present key areas for gully erosion control. Based on the visualization results and evaluation analysis conclusions, targeted gully erosion prevention and control measures and policy recommendations are proposed, providing a scientific basis for regional gully erosion control decisions.

[0114] 9. Optimization of measures for hill collapse control 9.1 Optimization of individual measures: Based on spatial interpolation data from Meizhou City, the parameters of intercepting ditches were optimized at the runoff collection points on the slopes, the height and slope of retaining walls were optimized at the bottom of the landslides, and the parameters of drainage projects were optimized in low-lying areas. For vegetation restoration, native broad-leaved tree species such as red pinna, Michelia champaca, and Liquidambar formosana were prioritized, and the planting density and timing were optimized to ensure coverage of the landslide slopes.

[0115] 9.2 Overall Optimization: Genetic algorithms were used to optimize the engineering and biological measures for gully collapse management in Meizhou City, taking into account both the management effect and cost.

[0116] 9.3 Optimization of Orchard Development Model: In line with the characteristics of Meizhou's leading agricultural industries, a land preparation and orchard development model was integrated into the management of landslides, achieving a win-win situation for ecological restoration and agricultural benefits. For landslide areas in the stable and semi-stable stages, pre-treatment methods such as slope cutting and land preparation, and soil improvement (adding organic fertilizer and improving the acidity of red soil) were implemented to remove debris, level the slopes, and create terraces or tiered fields suitable for fruit tree planting. Priority was given to selecting native economic tree species such as Meizhou pomelo and Sanhua plum, combined with native broad-leaved tree species such as red pine and Michelia champaca as orchard shelterbelts. This not only increased vegetation coverage and consolidated the management effect but also increased farmers' income through orchard development. Simultaneously, a complete orchard irrigation system was implemented, connecting with existing intercepting ditches and drainage projects to prevent secondary erosion caused by irrigation water accumulation. The planting row and tree spacing in the orchards was rationally planned, combined with the planting density requirements for vegetation restoration, to ensure that orchard development and ecological protection were promoted in tandem, adapting to the soil conditions and agricultural needs of Meizhou's red soil area.

[0117] 10. Application of the Economic Benefit Model for Governance of Hill Collapse To construct a unique industry-specific economic benefit model for Meizhou City, achieving a dual orientation of "ecological governance + economic benefits," the specific implementation steps are as follows: 10.1 Classification of Leading Industry Types Based on the industrial structure of various townships in Meizhou City, the townships where the collapse occurred are divided into three categories: agriculture-led, forestry-led, and tourism-led. An economic benefit model is constructed with a focus on agriculture-led townships to align with the characteristics of Meizhou City's agriculture-based industrial structure.

[0118] 10.2 Cost and Benefit Calculation Total cost accounting for remediation: Calculated over a 20-year remediation lifecycle, including one-time engineering construction costs, vegetation restoration costs, and annual monitoring and maintenance costs. The calculation formula is as follows: ,in, The total cost of governance over 20 years, Reduce construction costs by 10% in agriculture-led townships. Costs for vegetation restoration (including the cost of planting economic tree species). For the first Annual monitoring and maintenance costs.

[0119] Total benefits of water and soil conservation include direct benefits such as increased agricultural production and indirect benefits from water and soil conservation. The calculation formula is as follows: ,in, For the total benefits of governance, For direct income (agriculture-led townships focus on the increased production of cash crops such as Meizhou pomelos and Sanhua plums, tourism-led townships focus on the increased tourism revenue from increased visitor traffic to scenic spots, and forestry-led townships focus on the increased income from increased forest growth). Indirect benefits (the ecological benefits derived from red soil protection, water conservation, and soil erosion control).

[0120] 10.3 Economic Feasibility Assessment The economic feasibility of hill collapse remediation is assessed using three major economic indicators: Net Present Value (NPV), Internal Rate of Return (IRR), and Payback Period (Pt). Specific assessment standards for Meizhou City are developed, with a focus on lowering the standards for agriculture-dominated townships to better suit the characteristics of agricultural-based remediation benefits. The calculation formula is as follows: , ,in, For cash inflow, For cash outflow, The benchmark discount rate for Meizhou City is 6%. The accounting period is 20 years. This is the investment recovery period.

[0121] 10.4 Optimization of Governance Plan Based on the economic benefit model calculation results, targeted optimization of hill erosion control plans for townships with different industry types was implemented: agricultural-led townships increased the planting ratio of economic tree species such as Meizhou pomelos and Sanhua plums, and adopted a "company + farmer" model to enhance the added value of agricultural products; tourism-led townships combined hill erosion control with ecotourism to create water and soil conservation science parks and ecological landscape belts; forestry-led townships adopted a mixed "ecological forest + economic forest" model, and simultaneously applied for ecological compensation funds to supplement control costs, thereby maximizing ecological control and economic benefits.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the hazards of gully erosion, analyzing the urgency of treatment, and optimizing treatment, characterized in that, Includes the following steps: S1. Acquire meteorological data, topographic data, soil data, vegetation data, human disturbance data, and remote sensing imagery of the area where the gully collapsed, and perform preprocessing. S2. Identify extreme weather events in the area where the gully collapses, and establish a model of the correlation mechanism between the extreme weather events and the gully collapse erosion; S3. Construct an index system for assessing the intensity of human disturbance to evaluate the intensity of human disturbance in the area where the hill collapsed; S4. Extract gully collapse features from the preprocessed remote sensing images, verify the gully collapse features in the field, and conduct investigation and analysis in conjunction with the gully collapse micro-landforms; the gully collapse micro-landforms include catchment slopes, collapse walls, colluvial cones, gullies, and alluvial fans. S5. Based on the extreme climate events, the intensity of human disturbance, and the characteristics of gully collapse, construct an evaluation index system for the hazard of gully collapse erosion; S6. Construct a deep learning model that integrates convolutional neural networks, recurrent neural networks, and self-attention mechanism neural networks; based on the aforementioned ridge erosion hazard assessment index system and the aforementioned deep learning model, comprehensively evaluate the ridge erosion hazard and obtain the ridge erosion hazard assessment result; S7. Based on the assessment results of the hazards of gully erosion, a urgency assessment model for gully erosion control is constructed by combining multiple factors to obtain the urgency assessment results for gully erosion control; the multiple factors include at least the township's leading industry and the economic feasibility of control. S8. Using geographic information system technology, visualize the results of the hazard assessment of gully erosion and the results of the urgency assessment of gully erosion control, generate a grade distribution map, and propose targeted gully erosion control measures and suggestions based on the grade distribution map; S9. Based on the assessment results of the erosion hazards of the collapsed hills and the spatial interpolation data, a multi-objective optimization model with the optimization objectives of governance benefits and governance costs is constructed to optimize the parameters of the erosion governance measures. S10. Construct economic benefit models for hill collapse management under different leading industry types, and assess the economic feasibility of hill collapse management.

2. The method for evaluating the hazards of gully erosion, analyzing the urgency of treatment, and optimizing treatment according to claim 1, characterized in that, In S1, the meteorological data includes rainfall, rainfall intensity, temperature, and wind speed; The terrain data includes elevation, slope, and aspect; The soil data includes soil type, soil texture, and soil organic matter content; The vegetation data includes vegetation type and vegetation coverage; The human-caused disturbance data includes land use type and engineering construction activities; The remote sensing imagery includes high-resolution satellite imagery with a spatial resolution of no less than 0.5 meters and drone aerial imagery.

3. The method for evaluating the hazards of gully erosion, analyzing the urgency of treatment, and optimizing treatment according to claim 1, characterized in that, In S2, the establishment of a correlation mechanism model between extreme climate events and ridge erosion includes the following steps: Based on preprocessed meteorological data, extreme climate events are identified using extreme value theory and frequency analysis. Using the extreme climate events as input variables, a joint probability distribution model of extreme climate events and gully erosion is constructed using the Copula function, and the Markov chain method is used to simulate the evolution process of gully erosion under different extreme climate event conditions. The joint probability distribution model is used to analyze the impact mechanism of extreme climate events on the gully erosion process, and to quantify the contribution of the frequency, intensity and duration of extreme climate events to the amount of gully erosion; the contribution is used to construct the gully erosion hazard assessment index system.

4. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, is characterized in that, In S4, the characteristics of the landslide include morphological characteristics, erosion and sand production characteristics, and development stage characteristics; The morphological features include area, depth, volume, slope, perimeter, and aspect. The erosion and sediment yield characteristics include erosion modulus and sediment yield; The developmental stages are characterized by an active phase and a stable phase.

5. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, is characterized in that, In S5, the construction of the hazard assessment index system for gully collapse erosion includes the following steps: The subjective weights of the evaluation indicators are calculated using the analytic hierarchy process (AHP). The objective weights of the evaluation indicators are calculated using the entropy weight method. Based on the subjective weights and the objective weights, calculate the combined weights of the evaluation indicators; Grey relational analysis was used to calculate the correlation between evaluation indicators and the hazard of gully collapse and erosion, and indicators with a correlation below the preset threshold were removed. Principal component analysis is used to extract principal component factors whose cumulative variance contribution rate is greater than a preset threshold. Based on the loading matrix of the principal component factors, the retained evaluation indicators are clustered and reorganized to obtain an optimized evaluation indicator system.

6. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, is characterized in that, In S6, the construction of a deep learning model that integrates convolutional neural networks, recurrent neural networks, and self-attention mechanism neural networks includes the following steps: A convolutional neural network module was constructed to extract the spatial distribution and morphological features of the collapsed hills; A recurrent neural network module was constructed to extract the temporal variation features of extreme climate events and anthropogenic disturbances; A neural network module based on the self-attention mechanism is constructed to extract global correlation features among various evaluation indicators; The output features of the convolutional neural network module, the recurrent neural network module, and the self-attention mechanism neural network module are adaptively weighted and fused through a fusion layer, and then input into a fully connected layer to output the assessment result of the hazard of gully collapse erosion. The deep learning model is trained and optimized using ensemble learning and transfer learning methods. Sensitivity analysis and uncertainty analysis were used to verify the hazard assessment results of the gully collapse erosion.

7. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, is characterized in that, In S7, the construction of the urgency assessment model for addressing the collapse of jobs includes the following steps: Based on the assessment results of the hazard assessment of gully erosion, the frequency of extreme climate events, the intensity of human disturbance, regional ecological sensitivity, regional socio-economic importance, the leading industries of townships, and the assessment factors of the economic feasibility of governance, an assessment model for the urgency of gully erosion governance is constructed. Construct a judgment matrix and use the analytic hierarchy process (AHP) to calculate the subjective weights of each evaluation factor; The fuzzy comprehensive evaluation method is adopted, and the objective weight of each evaluation factor is calculated based on the subjective weight and the measured data of each evaluation factor. The combined weight of the evaluation factors is determined based on the subjective and objective weights. The urgency of addressing the collapse of farmland is divided into four levels: low, medium, high, and extremely high. Based on the urgency levels, a priority plan for addressing the collapse of farmland is formulated.

8. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, is characterized in that, In S9, the parameter optimization includes the following steps: Based on spatial interpolation data, determine the range of decision variables for the engineering measures parameter set and the biological measures parameter set; A multi-objective fitness function is constructed with the optimization objectives of maximizing governance benefits and minimizing governance costs. The decision variables are encoded using a genetic algorithm, and the Pareto optimal solution set is obtained by iteratively solving the multi-objective fitness function. The optimal parameter combination for the hill collapse control measures is determined based on the Pareto optimal solution set.

9. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 1, characterized in that, In S10, the construction of economic benefit models for the governance of job loss under different leading industry types includes the following steps: Based on the urgency assessment results of the gully collapse management and the optimized parameter combination of the gully collapse management measures, the type and scale of the gully collapse management engineering measures are determined; Obtain industrial structure data of the townships where the hills collapsed to determine the dominant industry types; By spatially overlaying the optimized parameter combinations with the dominant industry types, the potential impact range of governance measures on different industries can be determined. Calculate the total cost of remediation based on the determined type and scale of engineering measures; Based on the potential scope of impact and industry output data, predict the total benefits of governance; Based on the total cost and total benefit of governance, dynamic economic evaluation indicators are calculated; these dynamic economic evaluation indicators include net present value, internal rate of return, and investment payback period. The dynamic economic evaluation indicators are fed back to the multi-objective optimization model as constraints for parameter optimization.

10. The method for assessing the hazard of landslides, analyzing the urgency of remediation, and optimizing remediation according to claim 9, is characterized in that, It also includes analyzing the differences in the quantitative application of economic benefit models in different townships, including the following steps: Collect natural geographic data and socioeconomic statistics of the townships where the hills collapsed, and construct township feature vectors; The township feature vectors are input into the multi-objective optimization model. By adjusting the cost coefficient and benefit coefficient in the multi-objective optimization model, the governance costs and benefits of hill collapse management in different townships are calculated respectively. Establish a comparison matrix among townships that includes net present value, internal rate of return, and investment payback period; Based on the differences in the comparison matrix, key influencing factors affecting economic feasibility are identified. Based on the key influencing factors, the weights of the human disturbance intensity assessment index system are dynamically adjusted, or the optimal parameter combination of the collapse management measures is adjusted.

Images