Methods, equipment, media and products for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau

Abstract

The present application discloses a method, device, medium and product for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau, which relates to the research field of thermal melt disasters in the Qinghai-Tibet Plateau. The method includes: using the elevation, slope aspect, slope, terrain humidity index, annual average ground temperature, active layer thickness, solar radiation, normalized vegetation index, fine soil content and precipitation of thermal melt disaster points and non-thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau, training and optimizing a stacking model in which the primary learner is composed of a classification boosting tree, an extreme random tree, a random forest and an extreme gradient boosting tree, and the secondary learner is composed of a gradient enhanced decision tree, so as to obtain the prediction results of the predicted points at different geographical locations in the Qinghai-Tibet Plateau as thermal melt disaster points or non-thermal melt disaster points by using the optimized stacking model, and then obtain a thermal melt disaster susceptibility map for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau. The present application can improve the accuracy of the evaluation of the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

Description

Technical Field

[0001] The present application relates to the technical field of research on thermal melt disasters on the Qinghai-Tibet Plateau, and in particular to a method, equipment, medium and product for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau. Background Art

[0002] The Qinghai-Tibet Plateau, as the main area of ​​permafrost distribution in the world's mid- and low-latitude regions, is known as the "Third Pole of the Earth". This vast land is covered with about 1.06×10 6 Permafrost covers an area of ​​square kilometers, and its stability is extremely sensitive to climate change and environmental disturbances, making it a sensitive indicator of environmental change. In recent years, global warming has led to continuous degradation of permafrost in the Northern Hemisphere, and the warming rate on the Qinghai-Tibet Plateau is twice the global average, resulting in a significant rate of permafrost degradation, rising ground temperatures, increased active layer depth, and melting of ground ice. Permafrost degradation often triggers thermal melt disasters, which not only affect the landforms, but may also affect infrastructure, biogeochemical cycles, and change environmental conditions when they accumulate. With climate warming, the risks posed by thermal melt disasters and permafrost degradation are expected to increase. However, despite the widespread attention paid to thermal melt disasters, their potential risks to infrastructure and biogeochemical cycles are often overlooked, resulting in relatively few related studies. In order to assess the risk of regional thermal melt disasters under complex terrain conditions, geographic information systems (GIS) and remote sensing (RS) can be used to monitor the spatiotemporal changes in susceptible areas. Thermal melt susceptibility assessment (i.e., susceptibility evaluation) has been widely recognized as a key strategy for mitigating thaw subsidence disasters. Specifically, it requires evaluating the probability of disasters occurring in a specific area by analyzing local geological and environmental factors. Recent research progress has witnessed the growing application of machine learning in assessing the susceptibility of thermal melt disasters. However, current research usually uses a single machine learning model to learn the patterns of various environmental factors and obtain the assessment results of thermal melt disaster susceptibility, which has the problem of low accuracy in the evaluation of thermal melt disaster susceptibility.

[0003] In summary, the current methods for assessing the susceptibility of thermal melt disasters have the problem of low accuracy in assessing the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau. Summary of the invention

[0004] The purpose of this application is to provide a method, device, medium and product for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau, so as to improve the accuracy of the evaluation of the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a method for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau, the method comprising: obtaining a training data set; the training data set comprises a set of influencing factors of sample points at different geographical locations in the Qinghai-Tibet Plateau and a label marking each sample point as a thermal melt disaster point or a non-thermal melt disaster point; the set of influencing factors comprises elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation; using the training data set to train and optimize a stacking model to obtain an optimized stacking model; the stacking model comprises a primary learner and a secondary learner. The primary learner is composed of a classification boosting tree model, an extreme random tree model, a random forest model and an extreme gradient boosting tree model; the secondary learner is composed of a gradient enhanced decision tree model; a set of influencing factors of prediction points at different geographical locations in the Qinghai-Tibet Plateau is obtained; the set of influencing factors of each prediction point is respectively input into the optimized stacking model, and the prediction results of each prediction point being a thermal melt disaster point or a non-thermal melt disaster point are obtained by using the optimized stacking model; according to the prediction results of each prediction point being a thermal melt disaster point or a non-thermal melt disaster point, a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau is obtained; the thermal melt disaster susceptibility map is used for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

[0007] Optionally, the method for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau also includes: using Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effects to interpret the optimized stacking model, and analyzing the contribution rate of each influencing factor to the formation of thermal melt disasters within different numerical ranges through Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effects. When the contribution rate is positive, the influencing factor promotes the development of thermal melt disasters, and when the contribution rate is negative, the influencing factor inhibits the development of thermal melt disasters, and at the same time, the relationship between the influencing factors is obtained; the influencing factors are elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized vegetation index, fine soil content, and precipitation.

[0008] Optionally, before obtaining the training data set, the method for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau also includes: obtaining 10 influencing factors, including elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation; the correlation between any two of the 10 influencing factors is lower than 0.7; the correlation between any two of the 10 influencing factors is obtained using the Pearson correlation coefficient, variance inflation factor and tolerance evaluation; obtaining the influencing factors of thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-thermal melt disaster points; the number of thermal melt disaster points and the number of non-thermal melt disaster points are equal; and constructing a training data set using the influencing factors of thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-thermal melt disaster points.

[0009] Optionally, before obtaining the influencing factors of the thermal melt disaster points in different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of the non-thermal melt disaster points, the method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau also includes: trimming the 10 influencing factors of elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation to the entire Qinghai-Tibet Plateau and resampling to a spatial resolution of 1000 meters to obtain the influencing factors of the thermal melt disaster points in different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of the non-thermal melt disaster points.

[0010] Optionally, obtaining a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region based on the prediction results of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point specifically includes: obtaining a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region using ArcGIS software based on the prediction results of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point.

[0011] Optionally, the heat melt disaster susceptibility map is obtained by utilizing the natural breakpoint classification function of ArcGIS software to classify the prediction results of each predicted point as a heat melt disaster point or a non-heat melt disaster point into five susceptibility levels; the five susceptibility levels are extremely low, low, medium, high and extremely high.

[0012] Optionally, the method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau further includes: using a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau to assess the potential risks of key infrastructure, soil organic carbon and vegetation in the Qinghai-Tibet Plateau.

[0013] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for assessing the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau.

[0014] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned methods for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

[0015] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects.

[0017] The present application provides a method, device, medium and product for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau. A stacking model is used in which a primary learner is composed of a classification boosting tree model, an extreme random tree model, a random forest model and an extreme gradient boosting tree model, and a secondary learner is composed of a gradient enhanced decision tree model to obtain prediction results of prediction points at different geographical locations in the Qinghai-Tibet Plateau as thermal melt disaster points or non-thermal melt disaster points, and then a thermal melt disaster susceptibility map for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau is obtained. Since the stacking model is trained using the elevation, slope aspect, slope gradient, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized vegetation index, fine soil content and precipitation of thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau, as well as the elevation, slope aspect, slope gradient, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized vegetation index, fine soil content and precipitation of non-thermal melt disaster points, and the stacking model combines the predictions of multiple models, thereby improving the accuracy of the evaluation, thereby improving the accuracy of the susceptibility evaluation of thermal melt disasters on the Qinghai-Tibet Plateau. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A schematic flow chart of a method for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau provided in one embodiment of the present application.

[0020] Figure 2 Flowchart of the stacking model for this application.

[0021] Figure 3 This is a flow chart for heat-thaw hazard susceptibility assessment for this application.

[0022] Figure 4 This application is based on the explanation diagram of a single sample using SHAP.

[0023] Figure 5 Cumulative local impact diagram of the four main factors for this application.

[0024] Figure 6 A single sample explanation diagram based on LIME for this application.

[0025] Figure 7 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] The purpose of this application is to provide a method, device, medium and product for evaluating the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau, so as to improve the accuracy of the evaluation of the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau.

[0028] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0029] like Figure 1 As shown, the present application provides a method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau, comprising steps 101 to 105.

[0030] Step 101: Obtain a training data set; the training data set includes a set of influencing factors of sample points at different geographical locations in the Qinghai-Tibet Plateau and a label marking each sample point as a thermal melt disaster point or a non-thermal melt disaster point; the set of influencing factors includes elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation.

[0031] Before step 101, it also includes: obtaining 10 influencing factors, including elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation; the correlation between any two of the 10 influencing factors is less than 0.7; the correlation between any two of the 10 influencing factors is obtained using the Pearson correlation coefficient, variance inflation factor and tolerance assessment. Obtain the influencing factors of thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-thermal melt disaster points; the number of thermal melt disaster points is equal to the number of non-thermal melt disaster points. Use the influencing factors of thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-thermal melt disaster points to construct a training data set.

[0032] Among them, before obtaining the various influencing factors of thermal melt disaster points in different geographical locations in the Qinghai-Tibet Plateau and the various influencing factors of non-thermal melt disaster points, it also includes: clipping the 10 influencing factors of elevation, slope aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized vegetation index, fine soil content and precipitation to the entire Qinghai-Tibet Plateau and resampling to a spatial resolution of 1000 meters, and obtaining the various influencing factors of thermal melt disaster points in different geographical locations in the Qinghai-Tibet Plateau and the various influencing factors of non-thermal melt disaster points.

[0033] Step 102: Train and optimize the stacking model using the training data set to obtain an optimized stacking model; the stacking model includes a primary learner and a secondary learner; the primary learner is composed of a classification boosting tree model, an extreme random tree model, a random forest model, and an extreme gradient boosting tree model; the secondary learner is composed of a gradient enhanced decision tree model.

[0034] Step 103: Obtain a set of influencing factors for prediction points at different geographical locations in the Qinghai-Tibet Plateau.

[0035] Step 104: inputting the influencing factor set of each prediction point into the optimized stacking model respectively, and using the optimized stacking model to obtain the prediction result of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point.

[0036] Step 105: A thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region is obtained according to the prediction results of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point; the thermal melt disaster susceptibility map is used for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

[0037] This step 105 specifically includes: according to the prediction results of whether each predicted point is a thermal melt disaster point or a non-thermal melt disaster point, using ArcGIS software to obtain a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region.

[0038] Among them, the heat melt disaster susceptibility map is obtained by using the natural breakpoint classification function of ArcGIS software to classify the prediction results of each predicted point as a heat melt disaster point or a non-heat melt disaster point into five susceptibility levels; the five susceptibility levels are extremely low, low, medium, high and extremely high.

[0039] Specifically, the susceptibility assessment method for thermal melt disasters on the Qinghai-Tibet Plateau also includes: using Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effect to interpret the optimized stacking model, and analyzing the contribution rate of each influencing factor to the formation of thermal melt disasters within different numerical ranges through Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effect. When the contribution rate is positive, the influencing factor promotes the development of thermal melt disasters, and when the contribution rate is negative, the influencing factor inhibits the development of thermal melt disasters, and at the same time, the relationship between the influencing factors is obtained; the influencing factors are elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content, and precipitation.

[0040] Specifically, the method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau also includes: using the thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau to evaluate the potential risks of key infrastructure, soil organic carbon and vegetation in the Qinghai-Tibet Plateau.

[0041] The technical solution of the present application is described below with a specific embodiment.

[0042] This application provides a method for evaluating the susceptibility and risk analysis of heat melt disasters on the Qinghai-Tibet Plateau using an interpretable ensemble learning model. This method uses stacking models for the first time for evaluating the susceptibility and risk analysis of heat melt disasters. In order to solve the problem that the machine learning model has limited interpretability as a black box model, which poses a challenge to its application in disaster prediction and early warning systems, an interpretable machine learning method is used to improve model transparency and establish trust between users and models. For example, there are currently examples of using Shapley Additional Explanations (SHAP) to predict coronary heart disease mortality, using interpretable machine learning to estimate crop yields, and using Local Interpretable Model-Agnostic Explanations (LIME) to study traffic safety. In order to address the above limitations, this application uses an interpretable stacking algorithm to generate a reliable heat melt disaster susceptibility map (i.e., a heat melt disaster susceptibility map). In this application, firstly, historical thermal melt disaster locations and condition factors were collected, and the relationship between them was analyzed. Secondly, a stacking model was proposed and applied to obtain a reliable thermal melt disaster susceptibility map. In addition, a variety of interpretation methods were used to illustrate the decision-making process of the stacking model and analyze the key factors affecting thermal melt disasters. Finally, the potential risks of organic carbon (SOC), infrastructure, and vegetation were evaluated based on the susceptibility map obtained by the stacking model. These research results will help decision makers formulate management strategies that adapt to future environmental changes.

[0043] The Qinghai-Tibet Plateau, the research area of ​​this application, is well-known for its steep terrain and vast permafrost areas. Its permafrost area is approximately 1.06×10 6 square kilometers. From 1980 to 2018, the temperature, precipitation and soil moisture in the region have increased significantly. The average annual ground temperature (MAGT) of the Qinghai-Tibet Plateau is about 1.56°C, and the average active layer thickness (ALT) is 2.32 meters. The total area of ​​the Qinghai-Tibet Plateau is about 2.60×10 6 The Qinghai-Tibet Plateau is the highest plateau in the world, with an average altitude of about 4,000 meters. In addition, there are many lakes on the Qinghai-Tibet Plateau, of which the total area of ​​lakes with an area of ​​more than 1 square kilometer reaches 5×10 4 square kilometers, accounting for 1.67% of the total area of ​​the Qinghai-Tibet Plateau.

[0044] This application uses the heat melt disaster dataset for susceptibility assessment and potential risk analysis. The dataset catalog systematically collects spatial distribution information from existing datasets. This application screened a total of 136,187 heat melt disaster points that have been formed in the study area (i.e., the entire Qinghai-Tibet Plateau), providing a basis for subsequent research. In the analysis, this application randomly selected 10,000 heat melt disaster points for machine learning modeling (features are influencing factors, disaster points are labels, features are input into the model, and labels are predicted). In addition, an equal number of non-heat melt disaster points are randomly generated in non-heat melt disaster areas to ensure the robustness of susceptibility assessment, that is, this application randomly selected 20,000 sample points (including 10,000 heat melt disaster points and 10,000 non-heat melt disaster points) in different geographical locations in the Qinghai-Tibet Plateau. Then, the complete dataset consisting of positive samples (heat melt disasters) and negative samples (non-heat melt disasters) (including features and labels, features are influencing factors, and labels are positive and negative samples) is divided into training sets and test sets in a ratio of 7:3 for subsequent machine learning modeling. The modeling factors of the machine learning model are described in detail below. In addition to the data involved in the machine learning modeling process, the data of major infrastructure, SOC density and vegetation type are also used for potential risk analysis, that is, based on the susceptibility assessment results predicted by the stacking model, the major infrastructure, carbon storage, and vegetation type data are analyzed to further understand the potential risks that may be caused by thermal melt disasters. The infrastructure data comes from the National Geospatial Information Public Service Platform (https: / / www.tianditu.gov.cn / ). The SOC density and vegetation type data are obtained from relevant technical papers.

[0045] The modeling factors of the machine learning model, namely the influencing factors of the heat melt disaster susceptibility model, include elevation, aspect, slope, terrain moisture index, ground temperature (i.e., annual average ground temperature), active layer (i.e., active layer thickness), solar radiation, normalized difference vegetation index, fine soil content, and precipitation. The formation of heat melt disasters is affected by many factors. Based on the existing data and the understanding of the disaster formation mechanism, this application screened 10 key influencing factors for assessing the susceptibility of heat melt disasters. These influencing factors include elevation, aspect, slope, terrain moisture index, ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content, and precipitation. The above influencing factors were clipped to the study area and resampled to a spatial resolution of 1000 meters (i.e., the influencing factors were superimposed together and resampled to obtain sample values). Heat melt disaster is a geomorphic phenomenon caused by the melting of permafrost with high ice content. For permafrost on the Qinghai-Tibet Plateau, elevation is a key factor in determining the presence or absence of permafrost. The elevation data comes from the global digital elevation model (DEM). Aspect and slope are calculated again based on elevation data, which affect the permafrost thawing process and hydrological dynamics, respectively. The terrain wetness index is derived from public data and is used to quantify the impact of topography on hydrological processes. Regions characterized by high terrain wetness index values ​​usually exhibit abundant soil moisture, which is conducive to the formation of ground ice. Permafrost is a prerequisite for the generation of thermal thaw hazards. Ground temperature and active layer, as indicators of permafrost conditions, are obtained from public data with a spatial resolution of 1 km and a time range of 2000 to 2016. Solar radiation causes ground ice to melt, and the solar radiation factor is calculated based on the average of public data from 2000 to 2016. In permafrost areas, vegetation plays a vital role in regulating solar radiation, reducing soil temperature, and thus maintaining the stability of the permafrost layer. The normalized difference vegetation index is an annual average from 2000 to 2016 calculated using Google Earth Engine (GEE) based on three Landsat satellite datasets (Landsat5, 7, and 8). Fine soil content helps to accommodate excess ice, and soil data are obtained from public data. Precipitation has a significant impact on the thermal state of permafrost. Based on publicly available meteorological data, this application calculated the average annual precipitation accumulation between 2000 and 2016 to quantify the specific impact of precipitation on the thermal state of permafrost.

[0046] Detecting multicollinearity is a key step in susceptibility modeling because its presence may lead to bias and error in prediction results. By identifying and dealing with collinearity problems, the accuracy and reliability of the model can be improved. In this application, the Pearson correlation coefficient, variance inflation factor (VIF) and tolerance (TOL) are used to evaluate the correlation between the selected moderating factors (i.e., influencing factors) to achieve collinearity analysis. For the Pearson correlation coefficient, the correlation coefficient of each factor should follow the following criteria: |R|≤0.75, indicating the independence of each factor. For VIF and TOL, VIF is a measure to evaluate the relationship between the predictor variable and other variables. The higher the VIF value, the stronger the collinearity between the variable and other variables. TOL is the reciprocal of VIF. The lower the TOL value, the stronger the collinearity between the variable and other variables. VIF>10 or TOL<0.1 indicates that there is a multicollinearity problem and the variable needs to be eliminated. The correlation between the conditional factors (i.e., influencing factors) in this application is less than 0.7, that is, the correlation between any two factors is less than 0.7, which confirms that the 10 factors used in this analysis meet the premise of mutual independence. In addition, the results in Table 1 also confirm the applicability of the condition factors of the present application. In Table 1, precipitation refers to the amount of precipitation, and soil refers to the fine soil content.

[0047]

[0048] The modeling method of this application is introduced below.

[0049] Due to the complex causes of heat-melt disasters, the susceptibility assessment often involves multiple environmental factors. In recent years, the stacking algorithm has shown its effectiveness and achieved satisfactory results in assessing the impact of the diversity of environmental factors on susceptibility. This algorithm improves the accuracy and robustness of the assessment by combining the predictions of multiple models. Therefore, this application selects the stacking model to assess the susceptibility of heat-melt disasters. The architecture of the stacking algorithm consists of two parts: a primary learner and a secondary learner. The task of the primary learner is to extract key features, while the secondary learner uses these features for further analysis to draw the final assessment conclusion. In this application, the stacking model constructs a two-layer structure. The first layer consists of models such as random forest (RF), extreme random tree (EXT), classification boosting (CatBoost) and extreme gradient boosting (XGBoost), which are selected for their excellent performance and complex algorithmic principles. The second layer uses the gradient boosting decision tree (GBDT) as a secondary learner, which learns based on the output results of the first layer model to optimize the accuracy of the assessment. Among them, the classification boosting model is the classification boosting tree model, and the extreme gradient boosting model is the extreme gradient boosting tree model. The flowchart of the stacking model of this application is as follows Figure 2 As shown, Figure 2The middle extreme tree is the extreme random tree. Since the stacking algorithm combines the predictions of multiple models, that is, using the integrated learning framework to build the primary learner and the secondary learner, the primary learner is composed of RF, EXT, CatBoost, and XGBoost, and the secondary learner is composed of GBDT, the accuracy and robustness of the evaluation are improved.

[0050] The thermal melt susceptibility assessment process of this application is divided into four main steps, such as Figure 3 As shown, first, a data set containing thermal melt disaster events and their related conditional factors was established, that is, a complete data set consisting of positive samples (thermal melt disasters) and negative samples (non-thermal melt disasters), and the Pearson correlation coefficient, variance inflation factor and tolerance (TOL) were used to test the data set for multicollinearity (the results can be obtained by analyzing using SPSS software and Origin software). Secondly, the data set was randomly divided into 70% training sets and 30% test sets. Based on this, this application proposed a stacking model with the best performance, and used an integrated learning framework to construct a primary learner and a secondary learner. Finally, a stacking model was obtained, which consists of multiple machine learning algorithms, and the prediction accuracy of the model was evaluated, and the prediction results of other machine learning models were compared. In addition, in order to deeply understand the reasoning process of the model, this application adopted three interpretability methods: Shapley value additive explanation (SHAP), local interpretable model-agnostic explanation (LIME) and cumulative local effect (ALE). These methods help this application quantitatively analyze the impact of various regulatory factors on the formation of thermal melt disasters. The contribution rate of each influencing factor to the formation of heat melt disasters in different numerical ranges is analyzed by interpretable methods. If it is positive, it promotes the development of disasters, and if it is negative, it inhibits the development of disasters. At the same time, the relationship between the influencing factors can be obtained. Finally, this application uses the heat melt disaster susceptibility map (produced by ArcGIS software) to evaluate the potential risks of key infrastructure, soil organic carbon (SOC) and vegetation, that is, based on the susceptibility evaluation results predicted by the stacking model, the spatial distribution of major infrastructure, carbon storage, and vegetation type data is analyzed to further understand the potential risks that may be caused by heat melt disasters, and special attention is paid to the impact of high-risk areas (i.e. areas where heat melt disasters are concentrated in this section) on the Qinghai-Tibet Highway (QTH) and the Qinghai-Tibet Railway (QTR).

[0051] In the process of comparing model performance, this application uses a training data set (including features and labels, features are influencing factors, labels are positive samples and negative samples) for model training, and uses an independent test data set to calculate evaluation indicators. The assessment of thermal melt susceptibility is considered as a classification problem. For this purpose, this application uses accuracy ( ), accuracy ( )、Recall( )、F1 score( ) and the area under the receiver operating characteristic curve (AUC) are used as evaluation indicators. The selection of these indicators is consistent with the conventional practice of evaluating the performance of machine learning models in relevant technical literature. The specific calculation formulas are shown below.

[0052] .

[0053] .

[0054] .

[0055] .

[0056] Among them, TP (True Positive) and TN (True Negative) represent the number of correctly classified positive and negative samples, respectively, while FP (False Positive) and FN (False Negative) represent the opposite number. The receiver operating characteristic curve (ROC) and AUC values ​​are reliable indicators for accurately evaluating various disaster susceptibility models. The AUC value ranges from 0 to 1 and reflects the area under the ROC curve. In addition, the AUC value was selected as an additional indicator for quantitatively evaluating model performance. It classifies model performance into excellent (0.9-1), very good (0.8-0.9), good (0.7-0.8), moderate (0.6-0.7), and poor (0.5-0.6).

[0057] The following is an introduction to the model interpretability technology of this application.

[0058] Since machine learning models usually have black box-like characteristics, the thermal melt disaster susceptibility model is not interpretable enough, which may cause uncertainty and risk. It is difficult for users to thoroughly understand the specific mechanism of the model's predictive reasoning. In order to overcome this challenge, this application integrates explanatory techniques such as cumulative local effects, local interpretable model-agnostic explanations, and Shapley value additive explanations into the model, that is, three interpretability methods are used to analyze the trained model. The application of these methods is intended to enhance the transparency of the model, allowing users to clearly identify the specific impact of each input variable on the model's prediction results. The three interpretability methods used in this application are all independent of the model.

[0059] The cumulative local effect method was proposed by Daniel W. Apley and Jingyu Zhu to reduce the interference of inter-feature correlation on model prediction and accurately identify the independent impact of a single feature on the prediction result. The core of the ALE method is to quantify the local effect. It first sorts the target features and subdivides them into multiple small intervals, and determines a boundary value for each interval. On this basis, the feature value in each interval is set to its boundary value, while keeping other features unchanged, and then the prediction results under these boundary value conditions are calculated. Since the intervals are divided very carefully, this ensures that the values ​​of other features remain relatively stable during the evaluation process. In this way, the difference in prediction values ​​caused only by the change in the target feature value can be observed, and this change can be attributed to the target feature itself. ALE further accumulates all these local effects, providing a comprehensive assessment of the impact of each feature on the prediction result in the entire feature space. This accumulation process reveals the overall contribution of the feature to the prediction result, enhances the interpretability of the model and helps users understand the model decision process more deeply.

[0060] Local Interpretable Model-Agnostic Explanations is a model-independent algorithm proposed by Marco et al. The core goal of LIME is to provide easy-to-understand explanations for individual predictions of the classifier. The algorithm is based on a key assumption: for any given sample, the behavior of the classifier can be approximated by a simple local linear model. LIME works by introducing small perturbations to the input features in the neighborhood of the target data point. In this way, LIME observes how the model predictions change with these perturbations and assigns corresponding weights based on the similarity of the perturbed samples to the original samples. This weighting mechanism enables LIME to build an interpretable model for a specific local area, thereby generating highly interpretable predictions. In this way, LIME provides an intuitive and effective way to understand the decision-making process of complex models.

[0061] Shapley Additive Explanations (SHAP) is a widely used method in the field of machine learning, including deep learning, which focuses on the interpretability of model predictions. The SHAP method is based on the Shapley value principle in cooperative game theory, and assigns a numerical value to each feature in the model prediction, which quantifies the independent contribution of the feature to the prediction result. In this way, the importance of each feature to the model output can be deeply understood, while providing global and local explanations, thereby facilitating a comprehensive insight into the model behavior. In this application, the "shap" library in the Python 3.9.7 environment, especially the "KernelExplainer" function, was used to conduct an in-depth interpretability analysis of the stacked model. This analysis process helped this application reveal the internal working principles of the model and how different features work together to achieve the final prediction results.

[0062] The following is a detailed introduction to the research results of this application, including the relationship between heat melt disasters and influencing factors, model prediction and evaluation, the spatial distribution of heat melt disaster susceptibility maps, the interpretability of ensemble learning models, and the potential risks caused by heat melt disasters.

[0063] 1. The relationship between thermal melt disasters and influencing factors.

[0064] Histograms were used to explore the correlation between samples and ten conditional factors. This application analyzed the distribution percentages of heat melt disaster samples and non-heat melt disaster samples on each factor, and plotted the corresponding kernel density curves and average values, and obtained the following results: In terms of altitude, heat melt disasters tend to gather in the altitude range of 4381 to 5084 meters, accounting for 86.76%, and this altitude range is lower than the average altitude of non-heat melt disaster samples. Slope analysis shows that heat melt disasters and non-heat melt disasters have similar distribution characteristics on the north slope, northeast slope and south slope. In terms of slope, heat melt disasters are more common in gentle slope areas, and 92.64% of disasters occur in the slope range of 0 to 1.32 degrees, which is lower than the average slope of non-heat melt disaster samples. The terrain humidity index reflects the trend of water accumulation at any point in the basin, and heat melt disasters are more common in areas with higher terrain humidity index. Ground temperature has a significant impact on the frequency of heat melt disasters. As the ground temperature increases, the frequency of disasters shows a trend of first increasing and then decreasing, and its average ground temperature is higher than that of non-heat melt disaster areas. In terms of active layer thickness, heat melt disasters mainly occur in the range of 216.92 to 270.12 cm, with an average thickness of 228.78 cm. Solar radiation has a direct impact on the formation of heat melt landforms, and more than 81% of heat melt disasters occur in areas with solar radiation density of 181.87 to 190.90 watts per square meter. Analysis of the normalized vegetation index shows that more than 57% of heat melt disasters occur in areas with values ​​of 0.03 to 0.13. Data on fine soil content show that 76.41% of heat melt disasters are distributed in areas with values ​​of 51.12% to 67.56%. There is a complex relationship between precipitation and the frequency of heat melt disasters. With the increase in precipitation, the frequency of disasters shows a fluctuating trend of first rising, then falling, then rising, and then falling, and the average precipitation is higher than that in non-heat melt disaster areas.

[0065] 2. Model prediction evaluation.

[0066] Table 2 lists the performance of different models on various evaluation indicators in detail. Overall, the stacking model (the training data set prepared during model training is input, including features and corresponding labels, and the output is the label predicted by the model, and the training set and test set are divided in a ratio of 7:3) showed the best performance, followed by classification boosting, random forest, extreme gradient boosting, extreme random tree and gradient boosted decision tree. Specifically, the stacking model performed well in all key evaluation indicators, with an AUC of 0.9322, an accuracy of 0.8627, a recall rate of 0.9040, a precision of 0.8334, and an F1 score of 0.8673. CatBoost's performance is also quite close, with an AUC of 0.9316, an accuracy of 0.8605, a recall rate of 0.8932, a precision of 0.8367, and an F1 score of 0.8641. Random Forest has an AUC of 0.9286, an accuracy of 0.8568, a recall of 0.8781, a precision of 0.8406, and an F1 score of 0.8589. XGBoost has an AUC of 0.9276, an accuracy of 0.8563, a recall of 0.8845, a precision of 0.8357, and an F1 score of 0.8594. Extreme Random Trees has an AUC of 0.9275, an accuracy of 0.8600, a recall of 0.8872, a precision of 0.8398, and an F1 score of 0.8628. In comparison, GBDT performs slightly worse in the susceptibility evaluation, with an AUC of 0.9195, an accuracy of 0.8448, a recall of 0.8751, a precision of 0.8234, and an F1 score of 0.8484. The overall performance of the stacking model reached the highest level in terms of AUC and accuracy, which demonstrated its excellent predictive ability in thermal melt disaster susceptibility assessment.

[0067]

[0068] 3. Spatial distribution of thermal melt disaster susceptibility map.

[0069] After the model validation is completed, the next step is to produce a heat melt susceptibility map (i.e., heat melt susceptibility classification map), a tool that has been highly valued for its high utility in disaster risk management. This application integrates stacked models and individual models, such as classification boosting, random forest, extreme random tree (ET), extreme gradient boosting, and gradient boosted decision tree, into ArcGIS software to assess the potential for heat melt disasters within each grid cell. Furthermore, this application uses the natural breakpoint (Jenks) classification method to objectively divide the assessment results into five susceptibility levels: extremely low (i.e., very low), low, medium, high, and extremely high (i.e., very high). By using the natural breakpoint classification function of ArcGIS software for classification, the heat melt susceptibility classification map produced shows that the susceptibility assessment results of different models show consistency in spatial distribution, and the high susceptibility and extremely high susceptibility areas are mainly concentrated in the central region of the Qinghai-Tibet Plateau.

[0070] In order to fully evaluate the generalization ability of the model, this application uses a comprehensive dataset containing 136,187 heat melt disaster samples for validation, rather than using a smaller model dataset (originally containing 10,000 heat melt disaster samples). Tables 3, 4, and 5 list the feature distributions of different susceptibility categories in detail. In terms of the proportion of labeled high and extremely high susceptibility samples, the performance of the six models is as follows: the proportion of the stacking model is 20.08%, the classification boosting model is 19.76%, the extreme random tree model is 18.50%, the random forest model is also 18.50%, the extreme gradient boosting model is 17.70%, and the gradient boosting decision tree model is 19.80%. In the samples of high and extremely high susceptibility levels, the proportion of heat melt disaster points identified by the six models is: 91.20% for the stacking model, 89.24% for the ET model, 88.98% for the RF model, 88.96% for the CatBoost model, 87.68% for the XGBoost model, and 87.74% for the GBDT model. It is worth noting that the recognition rates of all models are over 87%, which indicates that all models show satisfactory performance, especially the stacked model, which has the highest recognition rate.

[0071]

[0072]

[0073]

[0074] 4. Interpretability of ensemble learning models (i.e., global interpretability of stacked models).

[0075] The additive interpretation of Shapley values ​​deeply analyzes the overall impact of each factor on the model prediction. For each factor, the corresponding SHAP value of the factor quantifies the specific impact of the factor on the prediction result. By sorting the total impact of these factors on the model prediction, the factors with the greatest impact, the factors with the least impact, etc. can be obtained, and the relative proportion of the importance of each factor can also be obtained. Among these factors, slope, elevation, terrain wetness index and precipitation are the four most critical features affecting the model prediction. It is worth noting that in the prediction training of thermal melt disasters, the orientation factor has the least impact on the model. The specific relationship between different factors and SHAP values ​​is: when the slope is low, its corresponding SHAP value is positive, which indicates that lower slopes (i.e. flat terrain) tend to promote the development of thermal melt disasters. Similarly, when the elevation, terrain wetness index and precipitation are at high levels, their SHAP values ​​are also positive, which means that the increase of these factors within a certain range promotes the formation of thermal melt disasters.

[0076] The additive interpretation of Shapley values ​​can comprehensively reveal the influence of each key factor in the data set on the development of heat melt disasters. Specifically, the relationship between factors such as slope, terrain moisture index, altitude and precipitation can be revealed in detail: when the slope is low, the SHAP value is positive, and the terrain moisture index is also high at this time, which indicates that the terrain is flat and the soil moisture content is high. These conditions jointly promote the formation of heat melt disasters. When the altitude is in the range of about 4500 to 5000 meters, the SHAP value is positive, indicating that this altitude range has a positive impact on the development of heat melt disasters. In addition, within this altitude range, the slope values ​​of most observation points are also low. High terrain moisture index and low slope value are favorable conditions for the development of heat melt disasters. When the precipitation is high, the SHAP value is also positive, which reflects the promoting effect of higher precipitation on the occurrence of heat melt disasters. There may be a nonlinear positive correlation between precipitation and terrain moisture index.

[0077] The SHAP summary plot provides a global interpretability view of the model and provides fine-grained local explanations of the prediction results for individual samples. Figure 4 This is an explanation diagram of a single sample based on SHAP for this application. Figure 4 Part (a) indicates thermal melt disasters. Figure 4 Part (b) represents non-thermal melt disasters. Figure 4 The sample features that affect the model's prediction of thermal melt disasters and non-thermal melt disasters are shown in detail, including features that increase the probability of non-thermal melt disaster prediction and features that increase the possibility of thermal melt disaster prediction. The length of the diagonal box intuitively reflects the degree of feature influence, and the black value in the figure represents the probability value predicted by the model. Figure 4Part (a) analyzes in depth the contribution of various characteristic factors in predicting heat melt disasters. In particular, slope, precipitation, altitude, terrain wetness index (TWI) and annual average ground temperature all contribute positively to the prediction results, among which the characteristic contribution of slope is the most significant, with a contribution value of 0.17. Figure 4 Part (b) shows that in the negative samples of non-thermal melt disasters, factors such as slope, TWI and altitude have a significant negative impact on the prediction results. Integrating SHAP values ​​into the machine learning workflow of this application enables in-depth analysis of the model's behavior, ensures the transparency of the model, and extracts valuable information from the data, thereby enhancing the credibility and practicality of the model.

[0078] The cumulative local effects method plays a dual role in the analysis: it aims to reveal trends in the forecast results and to exclude the interference of correlations between different factors, thereby more accurately highlighting the independent impact of each factor on the forecast results. This application specifically focuses on four key factors: slope, altitude, terrain wetness index and precipitation. Figure 5 The cumulative local impact diagram of the four main factors (i.e. key factors) of this application is shown in the figure below. Figure 5 Part (a) is the slope diagram. Figure 5 Part (b) is the elevation map. Figure 5 Part (c) is the TWI diagram. Figure 5 Part (d) in the middle is the precipitation map. Figure 5 Part (a) shows the ALE effect of slope on the predicted probability of thermal melt disaster in the stacking model. The results show that when the slope is less than about 5°, the predicted probability of thermal melt disaster decreases with the increase of slope. Figure 5 The elevation ALE analysis in part (b) reveals that the predicted probability of heat melt disaster is higher in the range of 4500 to 5000 meters above sea level. In addition, when the altitude is below 4500 meters, the average predicted value increases with the increase of altitude; while above 5000 meters, the average predicted value decreases with the increase of altitude. Figure 5 The ALE analysis of the terrain wetness index in part (c) shows that when the TWI value is greater than 0, it has a positive impact on the occurrence of thermal melt disasters, but this impact is not a monotonic linear relationship. Figure 5 Part (d) shows the ALE of the effect of precipitation on the formation of heat melt disasters. When the precipitation is between 100 and 500 mm, it is more conducive to the development of heat melt disasters.

[0079] The local interpretable model-agnostic explanation method aims to combine the prediction results of the simple model with the prediction results of the target prediction model by training a simplified model. This method is mainly used to analyze the interpretability of the prediction results of a single sample and provide a single sample explanation for the target prediction model. Figure 6This is a single sample explanation diagram based on LIME for this application. Figure 6 Part (a) indicates thermal melt disasters. Figure 6 Part (b) represents non-thermal melt disasters. Figure 6 The samples described in the LIME model correspond to the samples used in local interpretability in the SHAP model, ensuring that the results obtained from the two interpretable models are comparable. Figure 6 This paper illustrates the application of the LIME method to clarify the prediction results of stacked machine learning models. The horizontal axis depicts the weight that LIME assigns to each factor in the model, where a higher weight indicates a higher importance of the factor. A positive number indicates that the factor has a positive impact on the result, and a negative number indicates an opposite impact. Figure 6 The analysis in part (a) shows that the weight assigned to slope is 0.12, indicating that it is the most important, followed by TWI and elevation. Figure 6 In part (b), the weight of slope is -0.3, with the largest absolute value, indicating that it is the most important, followed by TWI, elevation, etc. In actual situations, thermal melt disasters are more likely to occur in permafrost areas with lower slopes and higher soil moisture content. The interpretation results of LIME technology are close to those of SHAP.

[0080] 5. Potential risks caused by thermal melt disasters.

[0081] It is crucial to assess the potential risks that may be caused by heat-melt disasters. Potential risk assessment based on superposition models (i.e. stacking models) can help identify which infrastructure areas are more vulnerable to such disasters and take effective prevention and mitigation measures accordingly.

[0082] In this application, a potential risk assessment was conducted for key infrastructures in the Qinghai-Tibet Plateau, including buildings, railway stations, railways, and highways, and relevant statistical data are provided in Table 6. The analysis results show that buildings and railway stations are relatively less affected, while road networks face greater threats of thermal thaw disasters. In terms of railways, about 388.12 kilometers of railway lines are located in high and extremely high susceptibility areas, most of which are distributed within the Qinghai-Tibet Engineering Corridor (QTEC). For highways, about 1024.22 kilometers of road sections are classified in high and extremely high susceptibility areas. Permafrost disturbance is a key factor affecting the development of thermal thaw landscapes, which is closely related to the emission of soil organic carbon. In addition, there are significant differences in the flux of greenhouse gases in thermal thaw landscapes under different vegetation types. In this application, a detailed statistical analysis of the distribution of soil organic carbon and vegetation types in high and extremely high susceptibility levels was conducted, and specific statistical results are provided in Tables 7 and 8. In the high susceptibility and very high susceptibility areas, the estimated average SOC density was 12.69 kg / m² and 12.64 kg / m², respectively. The SOC storage in these areas was estimated to be 1.32 PgC and 1.21 PgC, respectively, accounting for 19.06% of the total storage. In these susceptible areas, the main vegetation types are alpine steppe and alpine meadow, with distribution areas of 115,457 km² and 65,397 km², respectively.

[0083]

[0084]

[0085]

[0086] The Qinghai-Tibet Corridor is an important link from inland China to Tibet, and includes a number of key infrastructures on the Qinghai-Tibet Plateau. In the face of the potential threat of permafrost degradation in the region, which may affect the stability of infrastructure, this application has conducted a detailed system assessment. The analysis of this application is based on the stacking model results of the Qinghai-Tibet Railway and Qinghai-Tibet Highway selected within QTEC, while the infrastructure data comes from the Resource and Environmental Science Data Platform.

[0087] Assessing the potential risks associated with heat-thaw disasters is critical for identifying the vulnerable links of infrastructure, understanding their spatial distribution, and formulating corresponding management strategies. The potential risk assessment results based on the superposition model show that heat-thaw disasters are particularly concentrated in the central area of ​​the Qinghai-Tibet Plateau. In particular, within QTEC, the high-risk areas of QTR and QTH are mainly distributed in the area from Budongquan to Beilu River. According to the statistics in Table 9, the lengths of QTH sections at different susceptibility levels are: 70.71 km for extremely low susceptibility, 59.95 km for low susceptibility, 39.25 km for moderate susceptibility, 113.70 km for high susceptibility, and 222.45 km for extremely high susceptibility. The lengths of QTR sections corresponding to each susceptibility level are: 40.93 km for extremely low area, 67.05 km for low area, 52.15 km for moderate area, 122.83 km for high area, and 222.16 km for extremely high area. The lengths of high-prone and extremely high-prone areas of QTH and QTR are 336.15 km and 344.99 km respectively, accounting for 66.42% and 68.17% of the total length of QTEC. This data emphasizes the need to strengthen infrastructure monitoring and heat melt disaster management. Susceptibility assessment integrates multiple influencing factors and sample data sets, and uses expert knowledge or statistical methods to determine the weight of each factor, thereby gaining a deep understanding of the spatial distribution of the probability of heat melt disasters. Traditionally, previous studies have relied on training samples to evaluate the reliability of the results.

[0088]

[0089] In this application, additional heat melt disaster datasets from the Qinghai-Tibet Plateau Data Center (TPDC) that were not involved in model training were also used to verify the prediction accuracy of the model. The use of these additional datasets not only evaluates the generalization ability of machine learning models to handle nonlinear problems, but also tests the reliability of model predictions. Table 10 lists the training data statistics for various susceptibility categories used in this application, as well as the additional datasets of Niu and Luo (2022). More than 90% of the heat melt disaster points in both datasets are located in high susceptibility and extremely high susceptibility areas, confirming the reliability of the results. In addition, the results of this application are basically consistent with the research of Wang et al., with the proportion of dangerous points in high and extremely high susceptibility areas reaching 91.85%, slightly higher than the 90.61% reported by Wang et al.. In comparison with other datasets and similar studies, the results of this application show good reliability.

[0090]

[0091] Model selection is a crucial step in this application, but it poses a major challenge in the field of machine learning in the absence of clear guidelines. To address this challenge, this application adopts an innovative approach: a stacking model is introduced for the first time to assess the susceptibility of heat melt disasters and its real-time performance is compared with five other candidate models. The stacking model was selected based on its well-known ability to improve the accuracy of predictions by integrating multiple base models.

[0092] In the preliminary model evaluation, this application used a ten-fold cross-validation technique to test 11 popular machine learning models, as shown in Table 11. These models include various categories: CatBoost, extreme random trees, random forests, extreme gradient boosting, gradient boosted decision trees, adaptive boosting algorithm (AdaBoost), logistic regression (LR), Bayesian method (Bayes), classification and regression tree (CART), k nearest neighbor (KNN) and support vector machine (SVM). According to the test scores in Table 11, it can be found that strong learners usually score higher, while weak learners such as KNN score lower. Although SVM is a powerful learner, its score is relatively low, which may be due to its high requirements for data quality. In contrast, tree-based models are less stringent in terms of data requirements and can therefore achieve higher accuracy.

[0093] In further analysis, this application screened out the five best performing models (RF, EXT, CatBoost, XGBoost, and GBDT) and conducted iterative experiments to determine the best configuration for the stacked model. Finally, this application selected RF, EXT, CatBoost, and XGBoost as primary learners and GBDT as secondary learners. This combination fully utilizes the advantages of each model and produces excellent prediction results, as shown in Table 2.

[0094] Although the research in this application has achieved remarkable results in terms of accuracy, it also highlights the reliance on traditional models. Future research will explore more advanced methods, including developing new hybrid deep learning models, predicting the spatial distribution of heat melt disasters under future climate scenarios, and applying optimization algorithms to adjust model parameters to further improve model performance.

[0095]

[0096] This application focuses on the Qinghai-Tibet Plateau (QTP) and systematically collects data on various factors that affect the susceptibility of thermal melt disasters, including elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index (NDVI), fine soil content (FSC) and precipitation. These 10 factors constitute the criteria for susceptibility evaluation.

[0097] By applying the stacking model, this application conducted an in-depth assessment of the thermal melt susceptibility of QTP and achieved a high assessment accuracy. The main research findings of this application are summarized as follows.

[0098] (1) The stacking model was confirmed to be the best method for assessing the susceptibility of thermal melt disasters in QTP. The model showed excellent performance in all evaluation indicators, including AUC of 0.9332, accuracy of 0.8627, precision of 0.8334, recall of 0.9040 and F1 score of 0.8673, and its overall performance surpassed the other five machine learning models. The evaluation results based on the stacking model showed that about 20.08% of the permafrost areas in QTP were classified as high-susceptibility areas and extremely high-susceptibility areas, which covered 91.20% of the thermal melt disaster points.

[0099] (2) From the perspective of global interpretation, slope, elevation, TWI and precipitation are the most significant factors affecting the susceptibility of thermal melt disasters in QTP. In particular, areas with slopes less than 1.32 degrees, elevations between 4381 and 5084 meters, TWI between -0.49 and 1.36, and precipitation between 60.81 and 524.39 mm are the main areas prone to thermal melt disasters.

[0100] (3) The evaluation results of the stacking model show that the high-risk areas for thermal melt disasters in the QTP are mainly concentrated in the central region. Specifically, 388.12 kilometers of railways, 1024.22 kilometers of highways, 2.53 PgC of soil organic carbon storage, and 115,457 square kilometers of alpine grasslands are all located in high and very high risk areas. The Qinghai-Tibet Engineering Corridor is an area of ​​particular concern. Within the QTEC, about 336 kilometers of the Qinghai-Tibet Highway and 345 kilometers of the Qinghai-Tibet Railway were identified as high-risk areas, among which the potential risk of the section from Budongquan to Beilu River is particularly significant.

[0101] Compared with the common methods currently available, the method proposed in this application is more accurate, that is, the prediction accuracy is higher. While ensuring high accuracy, this application uses three interpretability methods independent of the model to conduct an interpretable analysis of the formation of thermal melt disasters, and deeply analyzes the key factors affecting its formation.

[0102] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the susceptibility assessment data of thermal melt disasters in the Qinghai-Tibet Plateau. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau is implemented.

[0103] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0104] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0105] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0106] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0107] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0108] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0109] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau, characterized in that: The method for assessing the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau includes: Acquire a training data set; the training data set includes a set of influencing factors of sample points at different geographical locations in the Qinghai-Tibet Plateau and a label marking each sample point as a thermal melt disaster point or a non-thermal melt disaster point; the set of influencing factors includes elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation; The stacking model is trained and optimized using the training data set to obtain an optimized stacking model; the stacking model includes a primary learner and a secondary learner; the primary learner is composed of a classification boosting tree model, an extreme random tree model, a random forest model and an extreme gradient boosting tree model; the secondary learner is composed of a gradient boosting decision tree model; Obtain the influencing factor set of prediction points at different geographical locations in the Qinghai-Tibet Plateau; Inputting the influencing factor set of each prediction point into the optimized stacking model respectively, and using the optimized stacking model to obtain the prediction result of each prediction point being a thermal melt disaster point or a non-thermal melt disaster point; According to the prediction results of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point, a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region is obtained; the thermal melt disaster susceptibility map is used for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau.

2. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 1, characterized in that: The method for assessing the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau also includes: The optimized stacking model is interpreted by using Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effect. The contribution rate of each influencing factor to the formation of heat melt disasters in different numerical ranges is analyzed by Shapley value additive interpretation, local interpretable model-unknowable interpretation and cumulative local effect. When the contribution rate is positive, the influencing factor promotes the development of heat melt disasters, and when the contribution rate is negative, the influencing factor inhibits the development of heat melt disasters. At the same time, the relationship between the influencing factors is obtained; the influencing factors are elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content, and precipitation.

3. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 2, characterized in that: Before obtaining the training data set, the method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau further includes: A total of 10 influencing factors were obtained, including elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation; the correlation between any two of the 10 influencing factors was less than 0.7; the correlation between any two of the 10 influencing factors was obtained using Pearson correlation coefficient, variance inflation factor and tolerance assessment; Obtain the influencing factors of thermal melt disaster points and non-thermal melt disaster points at different geographical locations in the Qinghai-Tibet Plateau; the number of thermal melt disaster points is equal to the number of non-thermal melt disaster points; The training data set is constructed using the various influencing factors of thermal melt disaster points in different geographical locations in the Qinghai-Tibet Plateau and the various influencing factors of non-thermal melt disaster points.

4. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 3, characterized in that: Before obtaining the influencing factors of heat-melt disaster points at different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-heat-melt disaster points, the method for evaluating the susceptibility of heat-melt disasters in the Qinghai-Tibet Plateau further includes: The 10 influencing factors, including elevation, aspect, slope, terrain moisture index, annual average ground temperature, active layer thickness, solar radiation, normalized difference vegetation index, fine soil content and precipitation, were clipped to the entire Qinghai-Tibet Plateau and resampled to a spatial resolution of 1000 meters. The influencing factors of thermal melt hazard points in different geographical locations in the Qinghai-Tibet Plateau and the influencing factors of non-thermal melt hazard points were obtained.

5. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 1, characterized in that: The prediction results of each predicted point being a thermal melt disaster point or a non-thermal melt disaster point, obtaining a thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau region specifically include: According to the prediction results of whether each prediction point is a thermal melt disaster point or a non-thermal melt disaster point, the thermal melt disaster susceptibility map of the Qinghai-Tibet Plateau was prepared using ArcGIS software.

6. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 5, characterized in that: The heat melt disaster susceptibility map is obtained by using the natural breakpoint classification function of ArcGIS software to classify the prediction results of each predicted point as a heat melt disaster point or a non-heat melt disaster point into five susceptibility levels; the five susceptibility levels are extremely low, low, medium, high and extremely high.

7. The method for evaluating the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to claim 1, characterized in that: The method for assessing the susceptibility of thermal melt disasters on the Qinghai-Tibet Plateau also includes: The potential risks of key infrastructure, soil organic carbon and vegetation in the Qinghai-Tibet Plateau were assessed using the heat melt susceptibility map of the Qinghai-Tibet Plateau.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau as described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for assessing the susceptibility of thermal melt disasters in the Qinghai-Tibet Plateau as described in any one of claims 1 to 7 is implemented.

Images