A dynamic assessment method for regional debris flow hazard based on spatiotemporal probability

Abstract

The present invention discloses a method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability, which belongs to the technical field of geological disaster prediction. The present application integrates meteorological data, historical disaster records and other data to establish a dynamic probability model in the time dimension and a distributed probability model in the space dimension, thereby comprehensively reflecting the characteristics of debris flow hazard changing with time and space. Specifically, the present application can solve the problem of result deviation caused by ignoring the temporal dynamic characteristics in traditional methods, and realize accurate prediction of debris flow hazard. In addition, the present application breaks through the limitations of traditional methods in data utilization, improves the accuracy and applicability of the model through the fusion of multi-source information, and significantly enhances the assessment and decision-making support capabilities for complex natural disasters.

Description

Technical Field

[0001] The invention belongs to the technical field of geological disaster prediction, and in particular relates to a method for dynamically evaluating regional debris flow hazard based on spatiotemporal probability. Background Art

[0002] Debris flow is a sudden and destructive geological disaster. Its hazard assessment is the core content of disaster prevention and risk management. In current research and application, the methods for assessing the hazard of debris flow mainly include traditional technologies based on statistical models, physical mechanisms, machine learning, and empirical judgment. These methods have played an important role in different research stages and application scenarios, forming a rich theoretical foundation and practical experience.

[0003] Although statistical models, physical mechanism methods, machine learning and empirical discrimination methods have their own advantages in debris flow hazard assessment, they still have significant shortcomings in practical applications and cannot fully meet the needs of dynamic assessment of complex natural disasters. These shortcomings can be analyzed from two aspects: the limitations of the model itself and the lack of data utilization.

[0004] The statistical model assumes that there is a linear relationship between environmental variables. Although this simplification is convenient for calculation, it is often difficult to hold true in the complex causes of debris flow.

[0005] Model limitations: Physical models usually ignore the complex interactions between environmental variables, and the computational complexity of the model also limits its application in large-scale areas. Machine learning methods can effectively capture nonlinear relationships, but the "black box" nature of the model makes it difficult for users to understand the specific contribution of variables to the results, thereby reducing its explanatory power in scientific research. At the same time, model training requires a large amount of high-quality sample data, and in areas with insufficient data, its prediction accuracy may drop significantly. In addition, the complexity of hyperparameter selection and the high demand for computing resources are also obstacles in practical applications. Empirical discrimination methods strongly rely on expert experience, lack systematicity, and are highly subjective. Especially in areas with complex environmental conditions or unknown disaster mechanisms, empirical discrimination methods are less applicable and difficult to provide accurate evaluation results.

[0006] Insufficient data utilization: Traditional methods are mostly based on a single data source and fail to fully integrate information in spatial and temporal dimensions. The risk of debris flow is affected by the combined effects of multiple factors, including climate conditions (such as rainfall intensity and frequency), topographic and geological characteristics, and changes in human activities. However, traditional methods usually adopt a static assessment framework and cannot dynamically capture the characteristics of these factors changing over time.

[0007] In summary, the main disadvantages of the traditional method are due to the following reasons:

[0008] The limitations of the theoretical basis of the method make it impossible to fully reflect the complex triggering mechanism of debris flows and the nonlinear interaction of environmental variables; the ability to integrate multi-source data is insufficient, and multi-dimensional information such as remote sensing and meteorology cannot be fully utilized; there is a lack of dynamic analysis capabilities, and it cannot reflect the impact of climate change and other time-related factors on the hazard of debris flows. These shortcomings restrict the applicability of traditional methods in complex dynamic environments. Summary of the invention

[0009] In order to solve the problems raised in the above background technology, the present invention provides a dynamic assessment method for regional debris flow hazard based on spatiotemporal probability. Traditional methods cannot fully reflect the complex triggering mechanism of debris flow and the nonlinear interaction of environmental variables, fail to make full use of multidimensional information such as remote sensing and meteorology, and fail to reflect the impact of climate change and other time-related factors on debris flow hazard.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] A method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability comprises the following steps:

[0012] S1: Collect watershed unit data and grid data of the area to be predicted. Each grid unit in the grid data includes its own historical meteorological data, future meteorological forecast data, material source distribution data, historical debris flow data and geographical data;

[0013] S2: By analyzing the relationship between historical debris flow data and historical meteorological data, the meteorological condition threshold value related to the occurrence of debris flow events in each grid unit is determined. The number of times the historical meteorological data in each grid unit exceeds the meteorological condition threshold value is counted, and the meteorological condition frequency of each grid unit is calculated. The time probability of debris flow events in each grid unit in the future is generated by the calculation method of future meteorological forecast data and meteorological condition frequency. Each watershed unit contains multiple grid units, so the time probabilities of all grid units in each watershed unit are weighted averaged to obtain the time probability P of debris flow events in each watershed unit in the future. t ;

[0014] S3: According to the source distribution data, set the source distribution conditions and set the location as ( x,y ) on the grid cell where the source exists indicator function M( x,y ), when the position is ( x,y ) grid unit meets the source distribution conditions, M( x,y ) takes the value of 1, otherwise it is 0. By x,y ) and the basin connectivity index I cPerform superposition calculation to obtain the spatial probability P of each watershed unit describing the possible impact range of debris flow s , Basin Connectivity Index I c It is a quantitative indicator that describes the material transfer efficiency between different geomorphic units in a watershed. The calculation method is:

[0015] ;

[0016] In the formula, D up is the possibility of sediment transport downward in the catchment area, D dn is the probability that the material will be transported to the nearest accumulation area via the water flow path, is the weight of the surface conditions of the catchment area, is the average slope of the catchment area, A is the catchment area, W i For the i The weight of the grid cells, S i For the i The slope of the grid cell, d i For the i The distance from the grid unit to the accumulation area, I c The value range of is (-∞, +∞);

[0017] Spatial probability P s The specific calculation method is:

[0018] The grid cells ( x , y ) of the material source existence indicator function M( x,y ) and watershed connectivity index I c ( x,y ) Use the raster calculator in ArcGIS to perform spatial overlay, filter out areas that meet the conditions, and set the threshold Determine which areas will become debris flow sources:

[0019] ;

[0020] Count the number of grid cells that meet the conditions within the watershed unit range, and estimate the total area of ​​the watershed based on the area of ​​the grid. P s That is, the proportion of grid cells that meet the conditions, the spatial probability of each watershed unit P s The calculation formula is expressed as:

[0021] ;

[0022] in, P s ( x,y ) is the spatial probability of the i-th grid cell, which is 1 if the condition is met, otherwise it is 0. A i is the area of ​​the ith grid cell, A total is the total area of ​​the entire study area;

[0023] S4: Take the disaster-preventing factors that cause debris flow disasters in geographic data as features, and use historical debris flow data to label "whether debris flow occurs". After constructing the training set and test set, the random forest model is obtained through training. Combined with the geographic data in the watershed unit, the debris flow susceptibility value P of each watershed unit is calculated. e ;

[0024] S5: The time probability P of each watershed unit t , spatial probability P s and the susceptibility value P e All of them are used as index values, and each index value is normalized to calculate the information entropy of each index. Then, the time probability P is calculated based on the information entropy. t , spatial probability P s and the susceptibility value P e The weight w t 、w s and w e , satisfying w t +w s +w e = 1, for basin unit j, calculate its comprehensive hazard index H according to the weighted formula j For: H j =w t ×P t +w s ×P s +w e ×P e , where H j The interval is [0,1].

[0025] Preferably, in S2, the meteorological condition threshold is set as:

[0026] T threshold ( i , j )=Percentile( T i，j , 96);

[0027] in,T threshold ( i , j ) represents the grid cell ( i , j ) meteorological condition threshold, T i，j Represents a grid cell ( i , j ) of historical meteorological data on debris flows, Percentile( T i，j , 96) represents the grid cell ( i , j )96% quantile value of historical meteorological data.

[0028] Preferably, in S2, for the grid unit ( i , j ), suppose there are N days of historical meteorological data, of which M days meet the threshold T threshold ( i , j ), then the grid cell ( i , j ) weather conditions frequency F i,j for:

[0029] F i,j =M / N。

[0030] Preferably, in S2, it is assumed that the future meteorological data has N future Days, among which M future If the meteorological conditions of the day reach the threshold, the grid cell ( i , j ) future time probability P t ( i , j )for:

[0031] .

[0032] Preferably, S5 specifically includes the following steps:

[0033] S5.1: Provided m There are three indicators for each watershed unit, namely, time probability P t , spatial probability P s and the susceptibility value P e , then:

[0034] ;

[0035] set up X i,j It is i The first j Index value;

[0036] S5.2: Normalize the values ​​of each indicator, that is, scale the value of each indicator to the interval [0,1]. The normalization formula is as follows:

[0037] ;

[0038] in, is the normalized value of the jth index value of the ith watershed unit, X j Represents the value of all watershed units of the jth indicator, max( X j ) and min( X j ) are respectively X j The minimum and maximum values ​​of

[0039] S5.3: Calculate the information entropy of each indicator, information entropy E j The calculation formula is:

[0040] ;

[0041] in, p i,j yes The ratio of the sum of the normalized values ​​of the jth index values ​​of all watershed units is expressed as:

[0042] ;

[0043] k As a constant, take Ensure that the entropy value range is between [0,1];

[0044] S5.4: Calculate the weight of each indicator based on information entropy w j , weight w j The calculation formula is:

[0045] ;

[0046] in, E j is the information entropy of the jth indicator;

[0047] S5.5: Substitute the required data through the above steps to calculate the time probability P t , spatial probability P s and the susceptibility value P e The weight w t 、w s and w e For each watershed unit j, the comprehensive risk H is calculated according to the weighted formula j :

[0048] H j =w t ×P t +w s ×P s +w e ×P e ;

[0049] Among them, H j The interval is [0,1].

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] This application integrates meteorological data, historical disaster records and other data to establish a dynamic probability model in the time dimension and a distributed probability model in the space dimension, thereby comprehensively reflecting the characteristics of debris flow hazards changing with time and space. Specifically, this application can solve the problem of result deviation caused by ignoring the dynamic characteristics of time in traditional methods, and achieve accurate prediction of debris flow hazards. In addition, this application breaks through the limitations of traditional methods in data utilization, improves the accuracy and applicability of the model through the fusion of multi-source information, and significantly enhances the assessment and decision-making support capabilities for complex natural disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the overall process of this application;

[0053] Figure 2 Schematic diagram of data composition and preprocessing;

[0054] Figure 3 This is a schematic diagram of time probability calculation;

[0055] Figure 4 This is a schematic diagram of spatial probability calculation;

[0056] Figure 5 Schematic diagram of debris flow susceptibility calculation. DETAILED DESCRIPTION

[0057] In order to facilitate those skilled in the art to understand the technical content of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] Embodiment 1;

[0059] like Figure 1 As shown, a method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability includes the following steps:

[0060] S1: Data preparation and preprocessing;

[0061] like Figure 2 As shown, the data used in this application include meteorological data, elevation data, satellite images and artificial remote sensing interpretation data. These data play different roles in the model establishment, calculation and optimization process. Since the model is highly reusable, subsequent applications only need to collect corresponding data to migrate the method to other areas, or replace data according to local geological and field conditions to ensure the adaptability of the method to local conditions.

[0062] Meteorological data: Historical meteorological data are from HRLT: A high-resolution (1 day, 1 km) and long-term (1961–2019) gridded dataset for temperature and precipitation across China (https: / / doi.org / 10.1594 / PANGAEA.941329), which is a gridded daily maximum temperature, minimum temperature, and precipitation dataset for China with a spatial resolution of 1 km and a time span of 1961-2019. The future meteorological data comes from NASA Global Daily Downscaled Projections, CMIP6 (https: / / doi.org / 10.1038 / s41597-022-01393-4). This dataset is based on the output of the CMIP6 model and has been statistically downscaled to provide a finer spatial resolution (0.25°×0.25°). Its time span is from 1950 to 2100, covering historical data and future scenario predictions, including key climate variables such as surface temperature (maximum and minimum) and precipitation, and supporting multiple SSP (Shared Socioeconomic Pathway) scenarios, such as SSP1-2.6, SSP2-4.5 and SSP5-8.5.

[0063] Digital elevation model: The benchmark digital elevation model (DEM) data used in the present invention is SRTMDEM (spatial resolution is 30m) provided by Google Earth data engine, which is used to extract terrain features (slope, curvature, etc.) and watershed analysis.

[0064] Remote sensing imagery: Remote sensing imagery data is an important basis for supporting the interpretation and factor extraction of debris flow disaster events. According to research needs and data acquisition conditions, remote sensing imagery data can be derived from a variety of channels, and commonly used data sources include Landsat, Sentinel, and Planet. Landsat data is provided by the United States Geological Survey, with a resolution of 30 meters (multispectral band) and 15 meters (panchromatic band). The time span is from Landsat1 in 1972 to the current Landsat8 and Landsat9, covering nearly 50 years of global imagery data. These data are very suitable for conducting spatiotemporal analysis and dynamic change research of historical debris flow events due to their long-term continuity and free characteristics. Sentinel-2 data provided by the European Space Agency is another important open source resource with a resolution of 10 meters (multispectral band) and a time span from 2015 to the present, with a revisit every 5 days. Sentinel data not only has a higher resolution, but also provides multi-band information, which is particularly suitable for disaster interpretation and surface factor extraction. The high-resolution images provided by Planet are a type of commercial remote sensing data with a resolution of 3 meters or even higher. The time span is short but the update frequency is very high, and some areas can be revisited daily. Its high resolution and high timeliness give it significant advantages in rapid response and refined analysis, especially in capturing surface changes before and after debris flow disasters. Combining the advantages of open source data and paid data, data sources can be flexibly selected according to actual needs. In large-scale regional assessments, Landsat and Sentinel data are more economical because they are free and have a long coverage period; for refined research in key areas and monitoring of dynamic changes in a short period of time, high-resolution Planet data can be introduced for supplementation and correction, thereby improving the accuracy and reliability of remote sensing interpretation.

[0065] Extraction of watersheds and main channels: Using the hydrological analysis function of the SWAT (Soil and Water Assessment Tool) model, input DEM data and complete a series of operations, including filling depressions, flow direction calculation, catchment analysis, confluence analysis, river network linking, river network classification, river network extraction, watershed division and vector output. Subsequently, the generated river network and watershed vector data are superimposed with terrain and remote sensing images to check their spatial distribution and accuracy. Through appropriate manual correction, the river network direction and watershed boundaries are revised, and finally high-precision watershed boundaries and main channel data are generated.

[0066] Collection of basic disaster-pregnant factors: The extraction of basic disaster-pregnant factors is an important part of debris flow disaster assessment, mainly including key factors in topography, geology, soil and land use. The acquisition and adaptation of these factors require the integration of multi-source data and reasonable adjustment in combination with regional characteristics to improve the scientificity and applicability of the factors. Topographic factors are the basic part of disaster-pregnant factors, extracted through digital elevation model (DEM), including slope, aspect, elevation, curvature, etc. These factors directly affect the convergence of water flow and slope stability, and are the core control factors for the occurrence of debris flow disasters. For example, the slope determines the driving force of material sliding, while the curvature affects the degree of water flow aggregation. When adapting the factors, it is necessary to select appropriate resolution and analysis scale according to the actual terrain characteristics of the study area. Soil factors involve soil permeability, particle composition, porosity and other characteristics, which determine the infiltration capacity of rainfall and the shear strength of soil. In terms of data sources, FAO's global soil database or regional soil survey data can be used. If the study area has a special soil structure, it is also necessary to verify it through experimental methods and adjust the classification standards to adapt to local conditions. Lithology factors directly reflect the geological background of the region, mainly including the composition of rocks, degree of weathering, and fault structure. These factors can be obtained through geological maps or remote sensing interpretation. If there are active fault zones in the region, a comprehensive analysis should be conducted in combination with seismic activity data. When adapting and adjusting, special attention should be paid to the spatial distribution and degree of fragmentation of lithology, which is of great significance for predicting potential debris flow prone areas. Land use and cover types affect vegetation coverage and surface hydrological processes, and indirectly control the risk of debris flow. Through remote sensing images (such as Sentinel-2 or Landsat series) and field surveys, land use classification information of the study area can be extracted, including forests, farmlands, grasslands, bare land, etc. It is necessary to regionalize the land use classification standards according to the actual situation of the study area, such as adding the classification of exposed slopes and impermeable areas related to debris flows. Adaptation and adjustment of factors is essential according to the unique landforms and environmental conditions of the study area. For example, the influence of glacial meltwater may need to be added in high-altitude areas; in arid areas, attention should be paid to the dynamic changes in soil moisture content. In addition, the interaction between factors is also an important basis for adjustment. For example, the vegetation coverage rate in high-slope areas may have a more significant effect on reducing debris flow disasters.

[0067] Debris flow disaster cataloging: Debris flow disaster cataloging is the basic work for conducting debris flow risk assessment and prevention and control research. Its core task is to clarify which river basins in the study area have experienced debris flow disasters, as well as the specific time, scale and impact range of the disaster. The cataloging work usually combines remote sensing interpretation, literature collection and data acquisition from geological disaster management departments to ensure the comprehensiveness and reliability of information. With the help of high-resolution remote sensing images (such as Sentinel-2, Planet, Landsat, etc.), by comparing images from different periods, typical characteristics of debris flow disasters can be identified, such as fan-shaped deposits, scouring marks, river blockage and landform destruction. Especially in the case of high image resolution, the morphological characteristics and range changes of debris flow accumulation areas can be clearly identified. For larger-scale disasters, terrain change analysis can also be combined with digital elevation models to determine the scale and intensity of the disaster. By consulting published academic papers, technical reports and local chronicles, recorded debris flow disaster events in the region can be obtained. These documents often contain background information, occurrence mechanisms and consequence analysis of disasters, and are important data sources to supplement remote sensing interpretation. In addition, relevant records may also be included in disaster prevention and mitigation plans, yearbooks and other materials issued by local governments. Geological disaster management departments (such as natural resources departments or geological survey agencies) usually have detailed debris flow disaster investigation and monitoring data, including information such as the location, time, causes and disaster losses of the disaster. These data are usually based on field investigations and long-term monitoring, and their high accuracy and reliability provide important support for disaster cataloging. Establishing cooperative relationships with relevant departments and obtaining the latest disaster cataloging data are also important links in compiling debris flow disaster catalogs.

[0068] Table 1 shows the expected results after completing all data processing steps:

[0069] Table 1 Summary of basin parameters;

[0070]

[0071] S2: time probability calculation;

[0072] In the assessment of debris flow hazard, probabilistic modeling in the time dimension is crucial, because the occurrence of debris flow is often closely related to specific meteorological conditions, and these conditions show obvious change characteristics in different time periods. In order to dynamically assess the risk of debris flow, it is necessary to consider the changes in meteorological conditions over time and the impact of such changes on the probability of debris flow occurrence. Therefore, constructing a probabilistic model in the time dimension can effectively capture and quantify the contribution of meteorological conditions in different time periods to the occurrence of debris flow, thereby providing a scientific basis for the spatiotemporal dynamic assessment of debris flow risk. Figure 3 As shown, the specific implementation process is as follows:

[0073] Meteorological condition threshold determination: By analyzing the relationship between historical debris flow data and historical meteorological data, the meteorological condition threshold that is closely related to the occurrence of debris flow is determined, and the frequency of occurrence of the meteorological condition threshold under past meteorological conditions is calculated. The data set of the present invention uses the 96% quantile as the meteorological condition threshold, which can reasonably reflect the extreme nature of meteorological conditions and match the frequency of historical debris flow events in the study area, thereby ensuring that the model can accurately predict possible debris flow events in the future.

[0074] For each grid cell, the 96% quantile of the meteorological data (such as precipitation, temperature, etc.) of the historical debris flow in the grid cell is calculated as the meteorological condition threshold for debris flow. The threshold is set as:

[0075] T threshold ( i , j )=Percentile( T i，j , 96);

[0076] in, T threshold ( i , j ) represents the grid cell ( i , j ) meteorological condition threshold, T i，j Represents a grid cell ( i , j ) of historical meteorological data on debris flows, Percentile( T i，j , 96) represents the grid cell ( i , j )96% quantile value of historical meteorological data.

[0077] Meteorological condition frequency calculation: After obtaining the meteorological condition threshold, count how many times the threshold condition is met in the historical meteorological data of each grid cell. That is, count the number of times the threshold is exceeded in the historical meteorological data and calculate its frequency. For the grid cell ( i , j ), assuming that there are N days of historical meteorological data, and the meteorological conditions of M days meet the meteorological condition threshold, then the meteorological condition frequency of the grid cell is F i,j for:

[0078] F i,j =M / N。

[0079] Time Probability Pt Generate; combine future meteorological forecast data with historical meteorological data, and calculate the probability of debris flow under future meteorological conditions by statistically analyzing the frequency of future meteorological conditions reaching the meteorological threshold. Assume that the time range of future meteorological data is T future , and in the future there will be N future Days, among which M future If the meteorological conditions of the day reach the threshold, then the future time probability of the grid cell P t ( i , j )for:

[0080] .

[0081] Statistics to watershed units; Statistics of the time probability of each grid unit to the watershed unit, assuming that the watershed unit X m Contains multiple grid cells ( i , j ), then the watershed unit X m The time probability P t ( X m ) is the weighted average of the temporal probabilities of all grid cells in the watershed.

[0082] Through the above steps, we can get the time probability of each watershed unit in the future P t ( X m ), which indicates the possibility that the meteorological conditions of the basin unit will reach the threshold for triggering debris flow in the future. Combined with the probability data of this time dimension, we can make a more accurate risk assessment of the occurrence of future debris flow and provide support for disaster warning and prevention.

[0083] S3: spatial probability calculation;

[0084] The spatial dimension probability of the present invention is based on the source distribution and the basin connectivity index to obtain the possible impact range of debris flow, such as Figure 4 As shown, the specific implementation process is as follows:

[0085] Provenance distribution extraction: If there is manually interpreted provenance distribution data, it will be used directly. If there is no manually interpreted data, the provenance distribution can be extracted in batches according to regional characteristics, such as slope greater than 15°; catchment area greater than a certain threshold (needs to be determined based on the hydrological characteristics of the study area); curvature greater than a certain threshold (such as positive curvature indicates convex characteristics, suitable for specific provenance distribution areas). GIS software or Python programming is used to calculate the above factors combined with digital elevation model data to generate a provenance distribution grid corresponding to the raster data;

[0086] In the data set used in this invention, the mathematical formula for extracting the source distribution can be expressed as:

[0087] ;

[0088] M( x,y ) is the grid position ( x,y ) is a source existence indicator function on the location, which is 1 if the location meets the source distribution conditions, otherwise it is 0.

[0089] Calculation of watershed connectivity index; Watershed connectivity index I c It is a quantitative indicator that describes the material transfer efficiency between different geomorphic units in a watershed. The common calculation method is:

[0090] ;

[0091] In the formula, D up is the possibility of sediment transport downward in the catchment area, D dn is the probability that the material will be transported to the nearest accumulation area via the water flow path, is the weight of the surface conditions of the catchment area, is the average slope of the catchment area, A is the catchment area, W i For the i The weight of the grid cells, S i For the i The slope of the grid cell, d i For the i The distance from the grid unit to the accumulation area, I c The value range of is (-∞, +∞).

[0092] Spatial probability P s Calculation; spatial probability P s The spatial range that debris flow may affect is expressed by superimposing the source distribution and the basin connectivity index:

[0093] First, the source existence indicator function M( x,y ) and the watershed connectivity index raster I c ( x,y ) to perform spatial overlay (using the raster calculator in ArcGIS), filter out areas that meet the conditions, and set thresholds To determine which areas will become debris flow sources:

[0094] ;

[0095] Based on experience or experimental data, the value used in this embodiment is 7.3.

[0096] Then, the number of grid cells that meet the conditions within the watershed unit is counted, and the total area of ​​the watershed is estimated based on the area of ​​the grid (usually related to the grid resolution). s The spatial probability P of each watershed unit is s The calculation formula is expressed as:

[0097] ;

[0098] in, P s ( x,y ) is the spatial probability of the i-th grid cell, which is 1 if the condition is met, otherwise it is 0. A i is the area of ​​the ith grid cell, A total is the total area of ​​the entire study area.

[0099] Through these steps, the spatial probability of describing the possible impact range of debris flow can be obtained. P s , providing a data basis for the spatiotemporal coupling model and further supporting comprehensive hazard assessment.

[0100] S4: Debris flow susceptibility modeling;

[0101] In the debris flow susceptibility modeling, random forest was selected as the main machine learning model. Random forest is an ensemble learning method that performs classification or regression by constructing multiple decision trees and combining their output results. Its advantages are high efficiency, robustness, and good adaptability to high-dimensional data. It is particularly suitable for processing complex, multi-dimensional geospatial data involved in debris flow susceptibility assessment. Figure 5 As shown, the specific implementation process is as follows:

[0102] The "basic disaster factors" in Table 1 are used as features, and the "whether debris flow occurs" in Table 1 is used as a label to construct a training set and a test set. 70% of the data is used for training and 30% of the data is used for testing. The model is evaluated using the AUC value. The AUC accuracy of the random forest model constructed by the present invention in the study area reaches 0.97. The random forest model obtained through training is combined with the spatial data in the region to calculate the debris flow susceptibility score P for each watershed unit. e These scores are used to generate a susceptibility distribution map, which can intuitively reflect the susceptibility of debris flow in the area.

[0103] S5: Comprehensive hazard calculation;

[0104] At present, the time probability P of each watershed unit has been obtained. t , spatial probability P s and the susceptibility value P e , the weights w of time probability, space probability and susceptibility in the basin hazard are determined by entropy weight method t 、w s and w e , satisfying w t +w s +w e =1. The implementation process of the entropy weight method is as follows:

[0105] ①Data preparation: Assume that there is m There are three indicators (temporal probability, spatial probability and susceptibility) for each watershed unit, then:

[0106] ;

[0107] set up X i,j It is i The first j indicator value; (e.g. i temporal probability, spatial probability or susceptibility of each watershed unit);

[0108] ② Normalization: In order to avoid the problem of different dimensions, each indicator value is first normalized, that is, the value of each indicator is scaled to the [0,1] interval. The normalization formula is as follows:

[0109] ;

[0110] in, is the normalized value of the jth index value of the ith watershed unit, X j Represents the value of all watershed units of the jth indicator, max( X j ) and min(X j ) are respectively X j The minimum and maximum values ​​of

[0111] ③ Calculate the information entropy of each indicator; information entropy reflects the uncertainty or dispersion of data. The more uniform the information, the higher the entropy, and vice versa. Information entropy E j The calculation formula is:

[0112] ;

[0113] in, p i,j yes The ratio of the sum of the normalized values ​​of the jth index values ​​of all watershed units is expressed as:

[0114] ;

[0115] k As a constant, take Ensure that the entropy value range is between [0,1];

[0116] ④ Calculate weights; calculate the weight of each indicator based on information entropy w j , weight w j The calculation formula is:

[0117] ;

[0118] in, E j is the information entropy of the jth indicator; the weight reflects the importance of each indicator. The larger the value, the greater the contribution of the indicator to the comprehensive evaluation.

[0119] ⑤ Calculate the comprehensive risk: For each watershed unit j, calculate its comprehensive risk H according to the weighted formula j :

[0120] H j =w t ×P t +w s ×P s +w e ×P e ;

[0121] Among them, H j The interval is [0,1].

[0122] ⑥ Hazard classification and visualization: The natural breakpoint method is used to automatically determine the dividing point according to the distribution of hazard values, and the hazard is divided into 5 intervals (such as very low, low, medium, high, and very high). j , add the attribute table of the watershed vector file, and use GIS software to generate a watershed hazard distribution map.

[0123] S6: Result verification and application:

[0124] In order to verify the accuracy of the debris flow hazard model, the present invention adopts a comprehensive verification method based on historical disaster data and field survey results. First, the high-risk areas predicted by the model are spatially overlapped with historical debris flow disaster records to check the consistency between the predicted results and the actual disaster location and scope. By comparing the impact areas of historical disasters with the hazard distribution output by the model, the accuracy of the model in spatial prediction is evaluated. At the same time, field survey data is also used to verify the reliability of the model, especially in areas where debris flow disasters have occurred. Through on-site inspections and post-disaster image comparisons, it is confirmed whether the high-risk areas predicted by the model are consistent with the actual disaster areas.

Claims

1. A method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability, characterized in that: The following steps are involved: S1: Collect watershed unit data and grid data of the area to be predicted. Each grid unit in the grid data includes its own historical meteorological data, future meteorological forecast data, material source distribution data, historical debris flow data and geographical data; S2: By analyzing the relationship between historical debris flow data and historical meteorological data, the meteorological condition threshold value related to the occurrence of debris flow events in each grid unit is determined. The number of times the historical meteorological data in each grid unit exceeds the meteorological condition threshold value is counted, and the meteorological condition frequency of each grid unit is calculated. The time probability of debris flow events in each grid unit in the future is generated by the calculation method of future meteorological forecast data and meteorological condition frequency. Each watershed unit contains multiple grid units, so the time probabilities of all grid units in each watershed unit are weighted averaged to obtain the time probability P of debris flow events in each watershed unit in the future. t ; S3: According to the source distribution data, set the source distribution conditions and set the location as ( x,y ) on the grid cell where the source exists indicator function M( x,y ), when the position is ( x,y ) grid unit meets the source distribution conditions, M( x,y ) takes the value of 1, otherwise it is 0. By x,y ) and the basin connectivity index I c Perform superposition calculation to obtain the spatial probability P of each watershed unit describing the possible impact range of debris flow s , Basin Connectivity Index I c It is a quantitative indicator that describes the material transfer efficiency between different geomorphic units in a watershed. The calculation method is: ； In the formula, D up is the possibility of sediment transport downward in the catchment area, D dn is the probability that the material will be transported to the nearest accumulation area via the water flow path, is the weight of the surface conditions of the catchment area, is the average slope of the catchment area, A is the catchment area, W i For the i The weight of the grid cells, S i For the i The slope of the grid cell, d i For the i The distance from the grid unit to the accumulation area, I c The value range of is (-∞, +∞); Spatial probability P s The specific calculation method is: The grid cells ( x , y ) of the material source existence indicator function M( x,y ) and watershed connectivity index I c ( x,y ) Use the raster calculator in ArcGIS to perform spatial overlay, filter out areas that meet the conditions, and set the threshold Determine which areas will become debris flow sources: ； Count the number of grid cells that meet the conditions within the watershed unit range, and estimate the total area of ​​the watershed based on the area of ​​the grid. P s That is, the proportion of grid cells that meet the conditions, the spatial probability of each watershed unit P s The calculation formula is expressed as: ； in, P s ( x,y ) is the spatial probability of the i-th grid cell, which is 1 if the condition is met, otherwise it is 0. A i is the area of ​​the ith grid cell, A total is the total area of ​​the entire study area; S4: Take the disaster-preventing factors that cause debris flow disasters in geographic data as features, and use historical debris flow data to label "whether a debris flow occurs". After constructing the training set and test set, the random forest model is obtained through training. Combined with the geographic data in the watershed unit, the debris flow susceptibility value P of each watershed unit is calculated. e ; S5: The time probability P of each watershed unit t , spatial probability P s and the susceptibility value P e All of them are used as index values, and each index value is normalized to calculate the information entropy of each index. Then, the time probability P is calculated based on the information entropy. t , spatial probability P s and the susceptibility value P e The weight w t 、w s and w e , satisfying w t +w s +w e = 1, for basin unit j, calculate its comprehensive hazard index H according to the weighted formula j For: H j =w t ×P t +w s ×P s +w e ×P e , where H j The interval is [0,1].

2. The method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability according to claim 1, characterized in that: In S2, the meteorological condition threshold is set as: T threshold ( i , j )=Percentile( T i，j ,96); in, T threshold ( i , j ) represents the grid cell ( i , j ) meteorological condition threshold, T i，j Represents a grid cell ( i , j ) of historical meteorological data on debris flows, Percentile( T i，j , 96) represents the grid cell ( i , j )96% quantile value of historical meteorological data.

3. The method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability according to claim 2 is characterized in that: In S2, for the grid cell ( i , j ), suppose there are N days of historical meteorological data, of which M days meet the threshold T threshold ( i , j ), then the grid cell ( i , j ) Weather Condition Frequency F i,j for: F i,j =M / N。 4. The method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability according to claim 3 is characterized in that: In S2, assume that the future weather data has N future Days, among which M future If the meteorological conditions of the day reach the threshold, the grid cell ( i , j ) future time probability P t ( i , j )for: 。 5. The method for dynamic assessment of regional debris flow hazard based on spatiotemporal probability according to claim 1, characterized in that: S5 specifically includes the following steps: S5.1: Provided m There are three indicators for each watershed unit, namely, time probability P t , spatial probability P s and the susceptibility value P e , then: ； set up X i,j It is i The first j Index value; S5.2: Normalize the values ​​of each indicator, that is, scale the value of each indicator to the interval [0,1]. The normalization formula is as follows: ； in, is the normalized value of the jth index value of the ith watershed unit, X j Represents the value of all watershed units of the jth indicator, max( X j ) and min( X j ) are respectively X j The minimum and maximum values ​​of S5.3: Calculate the information entropy of each indicator, information entropy E j The calculation formula is: ； in, p i,j yes The ratio of the sum of the normalized values ​​of the jth index values ​​of all watershed units is expressed as: ； k As a constant, take Ensure that the entropy value range is between [0,1]; S5.4: Calculate the weight of each indicator based on information entropy w j , weight w j The calculation formula is: ； in, E j is the information entropy of the jth indicator; S5.5: Substitute the required data through the above steps to calculate the time probability P t , spatial probability P s and the susceptibility value P e The weight w t 、w s and w e For each watershed unit j, the comprehensive risk H is calculated according to the weighted formula j : H j =w t ×P t +w s ×P s +w e ×P e ; Among them, H j The interval is [0,1].

Images