4.1 Distribution characteristics of factors affecting landslide points
This paper uses a visual interpretation of landslides by comparing Gaofen-1 and Jilin-1 satellite images before and after the Jieshishan Ms6.2 earthquake. Because there was a large amount of snowfall in the area before the earthquake, new soil was exposed in the area where the landslide occurred after the earthquake, which was conducive to the visual interpretation of satellite images of the landslide points. The principles of this visual interpretation are: Images with high spatial resolution in the preferred area are selected. If there are cloud cover or terrain shadows, images with similar time phases will be selected in order of spatial resolution from high to low, ultimately covering the entire earthquake zone. In the visual interpretation of landslide points, the main method used was to compare pre-earthquake and post-earthquake images. A total of 1,205 landslides and potential disaster points were catalogued in this interpretation (Fig. 2). Most of them were small collapses and landslides, mainly concentrated in the loess hilly areas on both sides of the Yellow River in the earthquake zone, near roads and valleys, and mostly developed on steep slopes of houses and roads 33. The main risk-bearing objects threatened were roads and farmland.
In order to further analyze the distribution characteristics of landslide points on various impact factors, the landslide points were superimposed on various factors in this study, and histograms were made for statistical analysis, where the horizontal axis was the classification of each factors and the vertical axis was the density.
For each Topographical factors (Fig. 3),In the analysis, the elevation factor was divided into nine levels at intervals of 100 m. The superimposed statistical analysis showed that the earthquake-induced landslide points were basically parabolic in elevation factor distribution (R2 = 0.7394), mainly distributed in the 1700–2250 m elevation zone; Regarding the distribution of landslide points on the slope, the slope was classified into 5° intervals. Statistics show that landslide points mainly occurred in the range of less than 7° and 20–25°, and were scattered in the range of slopes greater than 30°; On TWI, the occurrence of landslide points basically presents an exponential distribution (R2 = 0.6816); in terms of slope distribution, most of the earthquake-induced landslide disaster points occur in the east, southeast and south.
For various distance factors (Fig. 4), the distance factor from the road was classified into buffer zones with equal intervals of 0.5 km. The analysis found that the vast majority of landslide points induced by earthquakes occurred within 1.5 km of the road; For the distance from the fault zone, the first level is 1 km, and the second level and above are 2 km. The analysis shows that the earthquake-induced landslide points basically present an exponential distribution in the distance from the fault zone (R2 = 0.201), mainly distributed in the range of less than 1 km, 3–4 km and 8–12 km, among which the 3–4 km and 8–12 km intervals are on both sides of the Yellow River; In terms of distance from the river, the distribution of earthquake-induced landslide points is exponential (R2 = 0.7727), mainly distributed on both sides of the river and nearby; For the distance factor from the earthquake center, the first level in the study is counted as 5 km, and the second and subsequent levels are counted as 10 km. The analysis found that the landslide points are distributed exponentially (R2 = 0.9068).
At the same time, the distribution statistics of landslide points were also conducted based on Land use, Soil texture, NDVI and Population distribution factors (Fig. 5).In terms of Land use, earthquake-induced landslide points mainly occur in cropland, some grasslands, and some landslide also occur near water bodies. In terms of soil texture, earthquake-induced landslide points are mainly distributed in loam, with a small amount distributed in clay(light) and loam sand layers. In terms of NDVI, earthquake-induced landslide risks are mainly distributed between 0.08 and 0.16; Judging from the Population distribution in the earthquake-affected areas, human activities are more intense in places with high population density, and therefore the distribution of earthquake-induced landslide points is also correspondingly more.
4.2 Response analysis of impact factors on landslides
1) Model evaluation accuracy
The landslide points data and the selected 14 impact factors were input into the MaxEnt model. After 10 iterative calculations, the AUC value was finally obtained to be 0.854 (Fig. 6), and the model reliability reached a “good” level. Therefore, this study used the interpreted landslide points and various impact factors, and constructed the MaxEnt model through 10 iterative calculations to evaluate the risk of landslides induced by the Jishishan Ms6.2 earthquake. The results have good reliability.
2) Analysis of the importance of impact factors
Importance is an indicator that reflects the degree of model dependence on the variable 42. Table 5 shows the contribution rate and replacement importance of each impact factors to the impact degree of landslide disasters. It can be seen that the top five impact factors are distance from fracture zone, elevation, population distribution, soil texture data and distance from river. Their contribution rates were 39.0%, 38.1%, 17.8%, 1.3% and 1.2% respectively, and their cumulative contribution rate accounted for as high as 97.4%. As can be seen, The top five impact factors of replacement importance are distance from fracture zone, elevation, distance from river, population distribution and soil texture data, The replacement importance is 48.3%, 45.1%, 2.4%, 1.4% and 1.3% respectively, with a cumulative value of 98.5%.
Table 5
Contribution rate and replacement importance of the main disaster-causing factors
Serial No | Factors | Contribution rate/% | replacement importance/% |
1 | Distance from fracture zone | 39 | 48.3 |
2 | Elevation | 38.1 | 45.1 |
3 | Population distribution | 17.8 | 1.4 |
4 | Soil texture data | 1.3 | 1.3 |
5 | Distance from river | 1.2 | 2.4 |
6 | NDVI | 0.8 | 0.6 |
7 | Slope | 0.8 | 0.1 |
8 | Distance from road | 0.6 | 0.5 |
9 | Aspect | 0.3 | 0.2 |
10 | TWI | 0.1 | 0 |
11 | Land use | 0 | 0 |
12 | Sectional curvature | 0 | 0 |
13 | Flat curvature | 0 | 0 |
14 | Curvature | 0 | 0 |
Figure 7 shows the test results of the importance of each impact factors through the jackknife test method. From the test gain value 43 (Fig. 7a), it can be seen that the top five impact factors are distance from fracture zone, elevation, population distribution, soil texture data, and NDVI, with values of 0.35、0.32、0.22、0.12 and 0.1, respectively. According to the AUC values (Fig. 7b), the top five impact factors are elevation, distance to the fault zone, population distribution, NDVI, and distance to the river, with values of 0.74, 0.73, 0.67, 0.65, and 0.59, respectively. From the regularized training gain (Fig. 7c), it can be seen that the top five impact factors are elevation, distance to the fault zone, population distribution, distance to the road, and distance to the river, with values of 0.3, 0.29, 0.22, 0.1, and 0.08, respectively.
3) Analysis of the response of influencing factors to landslide risk
Figures 8 and 9 are the response curves of various impact factors to landslide occurrence, where the vertical axis represents the probability of landslide occurrence and the horizontal axis represents the value range of each factors. The reference probability threshold is set to 0.5. When it is greater than 0.5, it is considered that the value range of this factors is conducive to the occurrence of disasters 44. As shown in Fig. 8, the aspect has the highest response to landslide occurrence. A certain range of values of other factors is also sensitive to landslide occurrence. For example, when TWI is greater than 4m, the probability is greater than 0.5, which is very likely to cause landslides. Similarly, when the elevation zone is between 1700 and 2250 m, the profile curvature is -4.2 to 3, the plane curvature is -3.9 to 4.1, and the combined curvature is -6 to 11, this range responds strongly to landslides. When the distance to the fault zone is less than 1.7 km, the distance to the river is less than 3.8 km, the distance to the road is less than 2 km, the Slope is less than 30°, and the population distribution is less than 20 people/km2, the probability is greater than 0.5, which is very likely to cause landslides. It can also be seen that when NDVI is less than − 0.04 and 0.06 to 0.15, the probability is greater than 0.5, and the response to landslide disasters in this section is better; in terms of land use factors (Fig. 9), the probabilities of cultivated land, grassland and water areas are all greater than 0.5, and landslide disasters are very likely to occur; for soil texture, the probability of sandy loam and loam is greater than 0.5, and landslide disasters are very likely to occur.
4.3 Landslide risk assessment
This paper adopts the importance and correlation coefficient method of impact factors, calculates variance expansion factors test results method (Table 6), eliminates factors with strong collinearity (Planar curvature and Profile curvature), low contribution rate (Land use and Curvature) and correlation (Elevation), and then constructs a model with the remaining factors, calculates the maximum entropy results, and divides them into five levels according to the natural breakpoint method. Figure 10 shows the risk assessment results of the landslide induced by the Jishishan Ms6.2 earthquake obtained in this study. According to statistics, the area of extremely high risk zone is 49.38 km2, accounting for 0.84% of the total area of the study area; the area of high risk zone is 157.79 km2, accounting for 2.69% of the total area of the study area; the area of medium risk zone is 430.03 km2, accounting for 7.33% of the total area of the study area; the area of low risk zone is 526.07 km2, accounting for 8.96% of the total area of the study area; the area of extremely low risk zone is 4699.02 km2, accounting for 80.18% of the total area of the study area. It can be seen that since the earthquake occurred in winter, most places were seasonally frozen, so the landslides induced by this earthquake were mostly small, and the extremely high and high-risk landslides were relatively rare, mainly located in some areas on both sides of the Yellow River, which is consistent with the results of literature 33.
Table 6
Variance expansion factors test results
Impact factor | VIF |
Slope | 1.4 |
TWI | 1.26 |
Distance from road | 1.22 |
Distance from river | 1.13 |
NDVI | 1.12 |
Distance from fracture zone | 1.09 |
Aspect | 1.03 |
Soil texture data | 1.03 |
To further analyze the relationship between the risk zone and the earthquake intensity, the risk assessment results were superimposed on the earthquake intensity map 45, and the results were statistically obtained (Table 7). The density of extremely high and high risk zones is mainly located in the earthquake intensity VIII zone, with an area of 21.2 km2, accounting for 26.38% of the VIII zone. The density of medium risk areas is mainly distributed in VII and VIII zone, with an area of 341.22 km2, accounting for 16.92% and 28.82% of the area of VII and VIII zone respectively. The low and very low risk zone are mainly distributed in VII and VI zones, accounting for 75.33% and 97.55% of the area respectively. This area is far away from the earthquake-causing zone, and the risk of earthquake-induced geological disasters is also relatively low.
Table 7
Area percentage of different risk grades in different seismic intensity zones
earthquake intensity | extremely high risk/% | high risk/% | medium risk/% | low risk/% | very low risk/% |
VIII zone | 6.91 | 19.47 | 28.82 | 11.76 | 33.05 |
VII zone | 1.80 | 5.95 | 16.92 | 16.12 | 59.21 |
VI zone | 0.03 | 0.23 | 2.19 | 6.15 | 91.40 |