A method for evaluating landslide susceptibility in watershed based on weighted information model

Abstract

The present invention belongs to the technical field of geological disaster prediction and prevention, and relates to a method for evaluating the susceptibility of watershed landslides based on a weighted information model. The method comprises collecting landslide point data and multi-source environmental factor data in the target area, constructing a multi-source driving factor database, and performing preprocessing; calculating the explanatory power value of each factor using a geographic detector, and generating a dynamic weight coefficient through linear stretching; calculating the information value of each factor category using a traditional information model, and combining it with the obtained dynamic weight coefficient to generate a weighted information value; the present invention integrates natural and human factors to reveal the synergistic disaster-causing mechanism of "tectonic fragmentation-extreme precipitation-engineering disturbance"; analyzing the differences in driving mechanisms by watershed segment to improve the applicability of large-scale regional evaluation; compared with traditional IVM, the AUC value is increased from 0.795 to 0.875, and the misjudgment rate of extremely high-risk areas is reduced by 30%.

Description

Technical Field

[0001] The invention belongs to the technical field of geological disaster prediction and prevention, and relates to a method for evaluating the susceptibility of watershed landslides based on a weighted information volume model. Background Art

[0002] Landslide susceptibility assessment is a core component of geological disaster prevention and control. Traditional methods, such as the Information Volume Model (IVM), classify landslide risk based on the correlation between statistical factors and landslides. However, these methods fail to consider the spatial heterogeneity and weight differences between these factors, limiting prediction accuracy.

[0003] In the existing technology, patent (CN113420257A) discloses a landslide susceptibility assessment method based on machine learning, but does not solve the problem of dynamic adjustment of factor weights; patent (CN114841191A) discloses an epileptic EEG signal feature compression method based on a fully connected spiking neural network, which uses hierarchical analysis method for weighting, is highly subjective and difficult to quantify the interaction effects.

[0004] Landslides in large-scale regions, such as the Yangtze River Basin, are significantly influenced by the coupling of natural and human factors. There is an urgent need for an evaluation method that can integrate multi-source data, dynamically adjust weights, and reveal regional heterogeneity. This invention introduces a Geographical Detector to quantify the explanatory power of factors and, combined with an information content model, constructs a weighted evaluation system, significantly improving prediction accuracy and practicality. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method for evaluating the susceptibility of landslides in a watershed based on a weighted information model.

[0006] In order to achieve the purpose of the present invention, the present invention is implemented by adopting the following technical solutions.

[0007] A method for evaluating the susceptibility of watershed landslides based on a weighted information model includes the following steps:

[0008] S1. Collect landslide point data and multi-source environmental factor data in the target area, build a multi-source driving factor database, and pre-process the multi-source driving factor database;

[0009] The multi-source environmental factors include natural factors and human factors, wherein the natural factors include rainfall, slope, terrain relief, distance from faults, and distance from rivers; the human factors include the comprehensive human activity intensity index CHAII; wherein: the expression of the comprehensive human activity intensity index CHAII is:

[0010] ;

[0011] Where, NTL_AVG is the average nighttime light intensity at the landslide site, DOP_AVG is the average population density at the landslide site, and IC DIST is the distance between the landslide point and the impervious area and cultivated land; Norm() is a normalization function that normalizes the data to [0,1];

[0012] S2. Use the geographic detector to calculate the explanatory power q value of each factor and generate the dynamic weight coefficient P through linear stretching i ; Wherein: the explanatory power q value is expressed as:

[0013] ;

[0014] Among them, N h is the sample size of layer h, is the variance of layer h, is the overall variance;

[0015] The dynamic weight coefficient P i Expressed as:

[0016] ;

[0017] S3. Use the traditional information value model to calculate the information value I of each factor category i and the dynamic weight coefficient P obtained in step S2 i Combine to generate weighted information value ; Wherein: the information value I i Expressed as:

[0018] ;

[0019] in, is the landslide area within factor category i, S i is the total area of ​​factor category i, A0 is the total landslide area in the target area, and A is the total area of ​​the target area;

[0020] The weighted information value Expressed as:

[0021] ;

[0022] in, is the final weighted information value, which is used to characterize the comprehensive degree of regional landslide susceptibility.

[0023] As a preferred embodiment of the present invention, the preprocessing includes spatial normalization and factor discretization, wherein:

[0024] Spatial standardization, unified coordinate system and resolution, normalized to the range of 0,1;

[0025] Factor discretization uses geometric interval method or natural breakpoint method to classify continuous factors.

[0026] As a preferred solution of the present invention, the coordinate system is the WGS_1984_UTM_Zone_50N geographic coordinate system; and the resolution is a 100m×100m grid.

[0027] As a preferred embodiment of the present invention, the comprehensive human activity intensity index CHAII is obtained from human activity data.

[0028] As a preferred solution of the present invention, the human activity data includes night lights, population density, and land use; wherein the land use includes the distance between the landslide point and the impermeable area and cultivated land.

[0029] As a preferred solution of the present invention, the explanatory power q value is the q value of each factor to the landslide spatial density LSD.

[0030] As a preferred solution of the present invention, the dynamic weight coefficient P i It is obtained by calculating the explanatory power q value of each factor through normalized geographic detector.

[0031] As a preferred solution of the present invention, after dividing the target area into N segments, the above method is used to calculate the explanatory power q value of each factor in each segment and the interaction effect between different factors to identify the differences in key factors of each segment.

[0032] As a preferred solution of the present invention, the weighted information value obtained by the above method is used to divide the susceptibility levels using the natural breakpoint method to generate a watershed landslide risk zoning map.

[0033] As a preferred solution of the present invention, the accuracy of the weighted information value can be verified by the ROC curve and the AUC value.

[0034] Beneficial effects: Dynamic weight correction: quantifying factor contributions through geographic detectors overcomes the drawbacks of fixed weights in traditional models;

[0035] Multi-factor coupling analysis: integrating natural and human factors to reveal the synergistic disaster-causing mechanism of "tectonic fragmentation-extreme precipitation-engineering disturbance";

[0036] Adapting to regional heterogeneity: Analyzing driving mechanism differences by watershed segment to improve the applicability of large-scale regional assessments;

[0037] Significantly improved accuracy: Compared with traditional IVM, the AUC value increased from 0.795 to 0.875, and the misjudgment rate in extremely high-risk areas was reduced by 30%. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the present invention;

[0039] Figure 2 The ROC curve verification result of the traditional information model (AUC=0.80);

[0040] Figure 3 This is the ROC curve verification result of the weighted information model (AUC=0.88). DETAILED DESCRIPTION

[0041] The present invention will be further described with reference to the embodiments and the accompanying drawings.

[0042] As an embodiment of the present invention, Figure 1 As shown in FIG, a method for evaluating the susceptibility of watershed landslides based on a weighted information model includes the following steps:

[0043] S1. Collect landslide point data and multi-source environmental factor data in the target area, build a multi-source driving factor database, and pre-process the multi-source driving factor database;

[0044] The multi-source environmental factors include natural factors and human factors, wherein the natural factors include rainfall, slope, terrain relief, distance from faults, and distance from rivers; the human factors include the comprehensive human activity intensity index CHAII; wherein: the expression of the comprehensive human activity intensity index CHAII is:

[0045] ;

[0046] Where, NTL_AVG is the average nighttime light intensity at the landslide site from 2000 to 2020, DOP_AVG is the average population density at the landslide site from 2000 to 2020, and IC DIST is the average distance between the landslide point and the impervious area and cultivated land; Norm() is a normalization function that normalizes the data to [0,1];

[0047] S2. Use the geographic detector to calculate the explanatory power q value of each factor and generate the dynamic weight coefficient P through linear stretching i ; Wherein: the explanatory power q value is expressed as:

[0048] ;

[0049] Among them, N h is the sample size of layer h, is the variance of layer h, is the overall variance;

[0050] The dynamic weight coefficient P i Expressed as:

[0051] ;

[0052] S3. Use the traditional information value model to calculate the information value I of each factor category i and the dynamic weight coefficient P obtained in step S2 i Combine to generate weighted information value ; Wherein: the information value I i Expressed as:

[0053] ;

[0054] in, is the landslide area within factor category i, S i is the total area of ​​factor category i, A0 is the total landslide area in the target area, and A is the total area of ​​the target area;

[0055] The weighted information value Expressed as:

[0056] ;

[0057] in, is the final weighted information value, which is used to characterize the comprehensive degree of regional landslide susceptibility.

[0058] As an embodiment of the present invention, Figures 1 to 3 As shown in FIG, a method for evaluating the susceptibility of watershed landslides based on a weighted information model includes the following steps:

[0059] 1. Data Preparation: Integrate landslide data from 2000 to 2020, 30m DEM, rainfall raster, fault vectors, and CHAII data. These include: landslide event data from 2000 to 2020, with spatial duplication removed and outliers located, and kernel density analysis used to generate a spatial density distribution of landslides; 30m resolution digital elevation model (DEM) extraction of topographic factors such as slope and terrain relief; average rainfall raster data from 2000 to 2020 (unit: mm / year) to reflect the impact of regional rainfall on geological stability; geological fault and river vector data, with the Euclidean distance method used to calculate the distance of each pixel from the nearest fault and river landslide point flow; and the Human Activity Intensity Index (CHAII), which is constructed by integrating land use, nighttime light, and population density data to quantify the driving effect of human activities on landslide occurrence.

[0060] 2. Weight calculation: In order to avoid the problem of inconsistent influence of each factor on the model output, this step uses the geographic detector method to quantitatively analyze the explanatory power of each driving factor and perform weighted processing accordingly. The results show that the q value of the rainfall factor is 0.266, and the q value of the distance from the fault is 0.222. Both have significant explanatory power in landslide formation and are major controlling factors. Although the q value of CHAII in the single factor analysis is 0.188, there is an obvious enhanced interaction effect between it and other natural factors. Therefore, its weight is appropriately increased to 0.3516 during the weighted processing process to fully reflect the amplifying effect of human activities on the occurrence of landslides in local areas. Subsequently, the normalized q values ​​of each factor are introduced into the traditional information quantity model as weighting coefficients, and the model response accuracy is improved by the improved weighted information quantity (WIV) calculation method;

[0061] 3. Model Construction: After adjusting the weights of each factor, a grid overlay analysis of each evaluation factor was performed using the Weighted Information Volume Model (WIVM) to obtain a landslide susceptibility index map, which was then used to divide different risk levels. This step used the natural breakpoint method to divide the landslide susceptibility index into five levels: very low, low, medium, high, and very high. The results showed that extremely high-susceptibility areas are concentrated in areas of intense geological tectonic activity, such as the Three Gorges Reservoir and the western Sichuan Fault Zone, exhibiting typical spatial clustering characteristics and tectonic control effects. The WIVM more clearly distinguished between high-risk and low-risk landslide areas in space, avoiding the "overestimation" problem found in traditional models and effectively enhancing the model's practicality and disaster guidance capabilities.

[0062] 4. Verification and optimization: In order to evaluate the accuracy and improvement effect of the WIVM model, the landslide proportion and landslide ratio were used to verify the rationality of the landslide susceptibility model, and the ROC curve and AUC value were used as quantitative verification indicators. The results were compared with the traditional information volume model (IVM). The results in Table 1 show that the landslide proportions of the susceptible areas under each classification state are positively correlated from low susceptibility zones to high susceptibility zones, and the landslide ratios of the landslide hazard state classification under the two models meet the requirements of rationality test, that is, from extremely low susceptibility zones to extremely high susceptibility zones, the landslide proportion and landslide ratio (R nvalues) all showed a strictly monotonically increasing trend. The rationality test results demonstrated that both models could effectively distinguish landslide potential at different risk levels, meeting the core assumption of "high-risk areas with high density" in landslide susceptibility assessment and passing rationality verification. The ROC curve and AUC value verification results showed that the WIVM model achieved an AUC value of 0.875, a 10.1% increase compared to the IVM model's 0.795, significantly enhancing the model's ability to fit landslide distribution. Furthermore, the proportion of extremely high-risk areas decreased from 29.49% in the IVM model to 18.81% in the WIVM model, more closely matching the actual landslide distribution and reducing misjudgment of non-high-risk areas. Spatial comparative analysis of typical regions revealed that WIVM more accurately identified high-risk landslide areas in areas such as the Three Gorges Reservoir slopes and the western Sichuan fault zone, effectively improving the targeted and scientific nature of landslide early warning and disaster prevention. Figure 2 、 3 The comparison of ROC curves of IVM and WIVM models is shown.

[0063] Table 1:

[0064]

[0065] As an embodiment of the present invention, the preprocessing includes spatial normalization and factor discretization, wherein:

[0066] Spatial standardization primarily involves unifying the spatial reference system, grid resolution, and numerical range. To ensure spatial consistency across different data sources, all raster data was reprojected to the WGS 1984 coordinate system with a uniform resolution of 100 meters to match the spatial scale of the large-area model. For numerical standardization, the minimum–maximum normalization method was used to normalize each factor to the range [0, 1].

[0067] In terms of factor discretization, DEM-derived continuous variables such as slope, aspect, terrain relief, rainfall, and distance from faults were classified using the geometric interval method and the natural breakpoint method, respectively. The geometric interval method is suitable for variables with skewed distributions, such as rainfall and slope, and helps to retain the influence of extreme values; while the natural breakpoint method is more suitable for variables with quasi-normal distributions, such as terrain relief and distance from faults, and can more clearly delineate landslide susceptibility response zones. Taking the slope factor as an example, the original slope raster values ​​were standardized and divided into five levels using the natural breakpoint method, corresponding to extremely gentle, gentle, moderate, steep, and extremely steep slopes, effectively matching the critical slope characteristics of landslide activity.

[0068] As a preferred solution of the present invention, the coordinate system is the WGS_1984_UTM_Zone_50N geographic coordinate system; and the resolution is a 100m×100m grid.

[0069] As a preferred embodiment of the present invention, the comprehensive human activity intensity index CHAII is obtained from human activity data.

[0070] As an embodiment of the present invention, the human activity data includes night lights, population density, and land use; wherein the land use includes the distance between the landslide point and the impermeable area and cultivated land.

[0071] As an embodiment of the present invention, the explanatory power q value is the q value of each factor to the landslide spatial density LSD.

[0072] As an embodiment of the present invention, the dynamic weight coefficient P i It is obtained by calculating the explanatory power q value of each factor through normalized geographic detector.

[0073] As an embodiment of the present invention, before constructing the landslide susceptibility model, the study area was divided into three sub-regions according to geographical and administrative units (N=3 in this embodiment, divided by watershed units). Using the geographic detector method, the influencing factors in each region (such as rainfall, slope, terrain relief, distance from fault, CHAII, etc.) were calculated for factor explanatory power (q value) and two-factor interaction analysis. The analysis results show that there are significant differences in the dominant factors in different regions: for example, in the Three Gorges Reservoir area, CHAII interacts strongly with rainfall (the interaction q value is 0.412), while in the western Sichuan fault zone area, the distance from the fault and terrain relief are the dominant factors (q values ​​are 0.356 and 0.331, respectively). This step helps to identify regional key driving factors and realize zoning modeling and differentiated prevention and control of landslide risk zoning.

[0074] As an embodiment of the present invention, after obtaining the weighted information value (WIV) of each factor, a rasterized landslide susceptibility index layer was constructed. To enhance the scientificity and objectivity of the classification, the WIV values ​​were classified using the natural breakpoint method, categorizing the study area into five landslide susceptibility levels: very low, low, medium, high, and very high. The extremely high susceptibility areas are primarily concentrated in the steep slopes of the Three Gorges Reservoir area and near the active fault zone in western Sichuan; the medium-low susceptibility areas are mostly located in plains and areas less affected by human activities. The final landslide susceptibility zoning map was generated through visualization using the ArcGIS platform.

[0075] As an embodiment of the present invention, in order to verify the effectiveness and accuracy of the weighted information volume model (WIVM), the landslide proportion and landslide ratio (R n) for spatial validation. The landslide proportion refers to the proportion of landslide sample points in each risk level area, and the landslide ratio is the ratio of the number of landslides per unit area to the total landslide point proportion. The results show that the landslide proportion in the extremely high risk area reached 43.29%, and the landslide ratio was 2.30, which was significantly higher than that in other levels of areas, showing a good spatial aggregation. In addition, ROC and AUC values ​​were used to evaluate the overall classification ability of the model. Taking landslide points as positive classes and non-landslide points as negative classes, by comparing the predicted values ​​with the measured values ​​one by one, the final AUC value was 0.875, which was 10.1% higher than the traditional IVM model, verifying the reliability and superiority of the WIVM model in landslide susceptibility zoning. Figure 2 and Figure 3 shown.

[0076] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A method for evaluating the susceptibility of watershed landslides based on a weighted information model, characterized by: The steps include: S1. Collect landslide point data and multi-source environmental factor data in the target area, build a multi-source driving factor database, and pre-process the multi-source driving factor database; The multi-source environmental factors include natural factors and human factors, wherein the natural factors include rainfall, slope, terrain relief, distance from faults, and distance from rivers; the human factors include the comprehensive human activity intensity index CHAII; wherein: the expression of the comprehensive human activity intensity index CHAII is: ； Where, NTL_AVG is the average nighttime light intensity at the landslide site, DOP_AVG is the average population density at the landslide site, and IC DIST is the average distance between the landslide point and the impervious area and cultivated land; Norm() is a normalization function that normalizes the data to [0,1]; S2. Use the geographic detector to calculate the explanatory power q value of each factor and generate the dynamic weight coefficient P through linear stretching i ; Wherein: the explanatory power q value is expressed as: ； Among them, N h is the sample size of layer h, is the variance of layer h, is the overall variance; The dynamic weight coefficient P i Expressed as: ； S3. Use the traditional information model to calculate the information value I of each factor category i and the dynamic weight coefficient P obtained in step S2 i Combine to generate weighted information value ; Wherein: the information value I i Expressed as: ； in, is the landslide area within factor category i, S i is the total area of ​​factor category i, A0 is the total landslide area in the target area, and A is the total area of ​​the target area; The weighted information value Expressed as: ； in, is the final weighted information value, which is used to characterize the comprehensive degree of regional landslide susceptibility.

2. The method for evaluating the susceptibility of watershed landslides based on a weighted information model according to claim 1, wherein: The preprocessing includes spatial normalization and factor discretization, where: Spatial standardization, unified coordinate system and resolution, normalized to the range of 0,1; Factor discretization uses geometric interval method or natural breakpoint method to classify continuous factors.

3. The method for evaluating the susceptibility of watershed landslides based on a weighted information model according to claim 2, wherein: The coordinate system is the WGS_1984_UTM_Zone_50N geographic coordinate system; the resolution is a 100m×100m grid.

4. The method for evaluating the susceptibility of watershed landslides based on a weighted information model according to claim 1, wherein: The comprehensive human activity intensity index CHAII is obtained from human activity data.

5. The method for evaluating the susceptibility of landslides in a watershed based on a weighted information model according to claim 4, characterized in that: The human activity data includes night lights, population density, and land use; wherein the land use includes the distance between the landslide point and the impermeable area and cultivated land.

6. The method for evaluating the susceptibility of watershed landslides based on a weighted information model according to claim 1, wherein: The explanatory power q value is the q value of each factor on the spatial density of landslides.

7. The method for evaluating the susceptibility of watershed landslides based on a weighted information model according to claim 1, wherein: The dynamic weight coefficient P i It is obtained by calculating the explanatory power q value of each factor through normalized geographic detector.

8. The method for evaluating watershed landslide susceptibility based on a weighted information model according to claim 1, characterized in that: After dividing the target area into N segments, the method described in claim 1 is used to calculate the explanatory power q value of each factor in each segment and the interaction effect between different factors, and identify the differences in key factors of each segment.

9. The method for evaluating watershed landslide susceptibility based on a weighted information model according to claim 1, characterized in that: The weighted information value obtained by the method according to claim 1 is used to divide the susceptibility levels using the natural breakpoint method to generate a watershed landslide risk zoning map.

10. The method for evaluating watershed landslide susceptibility based on a weighted information model according to claim 1, characterized in that: The accuracy of the weighted information value can be verified by the ROC curve and AUC value.

Images