Essential insights into decision mechanism of landslide susceptibility mapping based on different machine learning models

Figures & data

Figure 1. Location and geological map of the study area.

Figure 2. Distribution and on-site investigation of landslides in the study area.

Table 1. Data and data sources.

Data name	Data sources	Type	Scale
Historical landslide	Chongqing Geological monitoring station	Dataset
Elevation	Aster satellite	Grid	30 m
Geological data	National Geological Data Center	Grid	1:200,000
Land cover	Chongqing Municipal Bureau of land and resources	Vector	1:100,000
Administrative division	Chongqing Municipal Bureau of land and resources	Vector	1:100,000
River network	Chongqing Water Resources Bureau	Vector	1:100,000
Annual rainfall	Chongqing Meteorological Administration	Dataset	90 m
Road	Chongqing Transportation Commission	Vector	1:100,000
Satellite image	Geospatial Data Cloud platform	Grid	30 m
POI of Chongqing	Web Crawler	Dataset

Figure 3. Flowchart of the study.

Table 2. Description of landslides conditioning factors.

Influence	Conditioning factor	Description
Topography	Elevation (m)	Elevation is the basic data for extracting terrain data, usually reflecting climate and vegetation patterns (Ngo et al. 2021).
	Aspect	Aspect is defined as the direction of projection of slope normal on horizontal plane. It plays an important role in mountain ecology. The orientation of mountain has an effect on sunshine hours and solar radiation intensity (Wang et al. 2019a).
	Slope (°)	Slope refers to the degree of surface unit steepness, equal to the vertical height of the slope divided by horizontal distance (Hong et al. 2019).
	Degree of relief (m)	It is a macroscopic index that describes the topographic characteristics of a region and characterizes the potential energy of surface erosion and material movement. In a certain area, the degree of relief is equal to the maximum value of elevation minus the minimum value. Different degree of relief regions have different influence on the occurrence of landslides (Wang et al. 2020).
	Curvature	Curvature can be seen as the slope of a slope. It stresses the materials on the slope and affects the movement of water on the slope surface (Chen et al. 2021; Moosavi and Niazi 2016).
	Plan Curvature	It is the slope perpendicular to the direction of the maximum slope. A positive value indicates that the cell is part of the lateral concave slope. A negative value indicates that the slope is laterally convex (Chen et al. 2021; Moosavi and Niazi 2016).
	Profile Curvature	The change rate of elevation along the maximum direction of slope becomes profile curvature, that is, the change rate of ground slope. A positive value indicates that the cell is part of a concave up slope. A negative value indicates that the slope is raised (Chen et al. 2021; Moosavi and Niazi 2016).
	Slope position	The core work of slope position extraction is the division of landform parts and landform types (Sun et al. 2021).
	Micro-landform	Relatively small-scale geomorphologic form and the smallest geomorphologic unit (Sun et al. 2020a).
	TWI	Based on DEM data, TWI comprehensively considers the influence of topography and soil on soil moisture distribution, and has important theoretical and application value in the study of spatial distribution of soil moisture in the basin (Sun et al. 2020c).
	TRI	TRI is the surface roughness index. Based on DEM, it is an index to measure surface fluctuation and erosion degree (Pham et al. 2020).
	STI	STI is a sediment transport index, which is used to characterize the surface sediment transport and deposition (Ali et al. 2021).
	SPI	The stream power index (SPI) is used to calculate the erosion value of surface water based on DEM data. The greater the value, the more obvious the erosion of surface water flow (Zhou et al. 2021).
Geological conditions	Distance from faults(m)	Setting a buffer zone to the fault, the closer to the main fault area, the looser the slope soil, more likely to cause landslides (Yi et al. 2020).
	Lithology	There are abundant outcropped strata and many types of lithology in the study area. Due to the different degrees of weakness of various strata lithology, their shear strength is different, and the difficulty of landslide is different (Wang et al. 2021).
	CRDS	CRDS refers to the relationship between the dip angle of rock and the aspect, and the direction of the sedimentary heap (Sun et al. 2020c).
Environmental conditions	Distance from hydrographic net (m)	The study area water system to establish buffers to determine the different scope of influence. Because the bank slope is cut and eroded by water flow, the slope closer to the river is more unstable, which leads to the occurrence of landslides (Kavzoglu et al. 2015).
	NDVI	Vegetation growth is another key factor reflecting slope stability. Therefore, the normalized difference vegetation index (NDVI) was used to detect the vegetation growth status, vegetation coverage and eliminate some radiation errors (Chen et al. 2021).
	Land cover	Land cover refers to the surface complex covered by natural and artificial buildings, including vegetation, soil, lakes and various buildings, with different land stability (Wang et al. 2019a).
	Annual average rainfall (mm)	The data are derived from the average rainfall from 2000 to 2014 to reflect the long-term rainfall and distribution characteristics in the study area (Sun et al. 2022b).
Human activities	Distance from roads (m)	The construction of a large number of roads inevitably poses a threat to the stability of the slope and is easy to induce landslides (Sun et al. 2022a).
	POI kernel density	POI nuclear density covers almost all types of human engineering buildings. Similar to the construction of highways, these buildings will also loose the soil around the slope, thus affecting the slope stability (Wang et al. 2021).

Table 3. Classification of landslide-conditioning factors.

Conditioning factor	Classification standard
Elevation (m)	1. <311; 2. 311–496; 3. 496–667; 4. 667–831; 5. 831–993; 6. 993 ∼ 1157; 7. 1157–1489; 8. 1489–1676; 9. 1676–1886; 10. 1886–2130; 11. >2130
Slope (°)	1. < 5; 2. 5–10; 3. 10–15; 4. 15–20; 5. 20–25; 6. 25–30; 7. 30–35; 8. 35–40; 9. >40
Degree of relief (m)	1. <20; 2. 20 ∼ 30; 3. 30–40; 4. 40–50; 5. 50–80; 6. 80–170; 7. >170
Aspect	1. Flat; 2. North; 3. Northeast; 4. East; 5. Southeast; 6. South; 7. Southwest; 8. West; 9. Northwest
Slope position	1. Ridge; 2. Upper slope; 3. Middle slope; 4. Flats slope; 5. Lower slope; 6. Valleys;
Micro-landform	1. Canyons, and deeply incised streams; 2. Mid-slope drainages, and shallow valleys; 3. Upland drainages, and headwaters; 4. U-shape valleys; 5. Plains; 6. Open slopes; 7. Upper slopes, and mesas; 8. Local ridges hills in valleys; 9. Mid-slope ridges, and small hills in plains; 10. Mountain tops, and high narrow ridges
Curvature	1. <-1; 2. −1–−0.5; 3. −0.5–0; 4. 0–0.5; 5. 0.5–1; 6. > 1
Profile curvature	1. <-1; 2. −1–−0.5; 3. −0.5–0; 4. 0–0.5; 5. 0.5–1; 6. > 1
Plan curvature	1. <-1; 2. −1–−0.5; 3. −0.5–0; 4. 0–0.5; 5. 0.5–1; 6. > 1
Lithology	1. D, C; 2. E; 3. J; 4. O; 5. P; 6. Q4; 7. S; 8. T1; 9. T2; 10. T3; 11. Others
Distance from faults (m)	1. < 500; 2. 50–1000; 3.1000–1500; 4. 1500–2000; 5. 2000–2500; 6. 2500–3000; 7. > 3000
CRDS	1. Dip-slope I; 2. Dip-slope II; 3. Outward slope; 4. Oblique slope; 5. Tangential slope; 6. Reverse slope; 7. Flat
TRI	1. < 1.05; 2. 1.05–1.1; 3. 1.1–1.15; 4. 1.15–1.2; 5. > 1.2
TWI	1. < 4; 2. 4–6; 3. 6–8; 4. 8–10; 5. >10
STI	1. < 20; 2. 20–40; 3. 40–70; 4. 70–100; 5. 100–200; 6. > 200
SPI	1. < 15; 2. 15–30; 3. 30–45; 4. 45–60; 5. 60–100; 6. 100–1000; 7. > 1000
NDVI	1. <0; 2. 0–0.1; 3. 0.1–0.2; 4. 0.2–0.3; 5. 0.3–0.4; 6. >0.4
Distance from hydrographic net (m)	1. < 100; 2. 100–200; 3. 200–300; 4. 300–500; 5. 500–600; 6. > 600
Annual average rainfall (mm)	1. <1157; 2. 1157–1202; 3. 1202–1244; 4. 1244–1286; 5. 1286–1329; 6. 1329–1376; 7. >1376
Land cover	1. Meadow; 2. Farmland; 3. Water area; 4. Forest; 5. Orchard; 6. Transportation land; 7. Urban land; 8. Others
Distance from roads (m)	1. <100; 2. 100–200; 3. 200–300; 4. 300–500; 5. 500–600; 6. >600
POI kernel density	1. 0–0.5; 2. 0.5–1; 3. 1–2; 4. 2–3; 5. 3–10; 6. >10

Figure 5. BP neural network model structure.

Figure 6. The schematic diagram of the random forest (RF) algorithm.

Figure 7. Landslide susceptibility mappings: (a) ANN model; (b) RF model.

Table 4. The accuracy of 10-fold cross-validation.

Subset (ANN)	Accuracy		Subset (RF)	Accuracy
	Training dataset	Test dataset		Training dataset	Test dataset
1	0.967	0.884	1	1.000	1.000
2	0.949	0.920	2	1.000	0.995
3	0.945	0.948	3	1.000	0.994
4	0.972	0.855	4	1.000	1.000
5	0.951	0.914	5	1.000	0.990
6	0.970	0.839	6	1.000	0.995
7	0.990	0.902	7	1.000	0.990
8	0.941	0.827	8	1.000	1.000
9	0.972	0.919	9	1.000	1.000
10	0.942	0.873	10	1.000	1.000

Table 5. Statistics of the susceptibility classes.

	Class:	Very low:	Low:	Moderate:	High:	Very high:
	Coverage (%)	50 %	20 %	10 %	10 %	10 %
ANN	Landslide (number)	223	124	113	173	233
	Landslide (%)	26 %	14 %	13 %	20 %	27 %
	Landslide density (Pcs/km^2)	0.145	0.234	0.461	0.605	0.700
RF	Landslide (number)	51	98	86	175	456
	Landslide (%)	6 %	11 %	10 %	20 %	53 %
	Landslide density (Pcs/km^2)	0.035	0.167	0.293	0.596	1.552

Figure 8. ROC curve of the ANN and RF models: (a) ANN model; (b) RF model.

Table 6. Confusion matrix of the ANN and RF models.

Model			Actual value
			Non-landslide (0)	Landslide (1)
ANN	Predicted value	Non-landslide (0)	822	37
		Landslide (1)	44	829
RF			Non-landslide (0)	Landslide (1)
	Predicted value	Non-landslide (0)	866	0
		Landslide (1)	0	866

Figure 9. Factor importance ranking: (a) ANN; (b) RF.

Figure 10. Selection of case units: (a) ANN; (b) RF.

Figure 11. Waterfall plot of landslide case unit: (a) ANN; (b) RF.

Figure 12. Waterfall plot of non-landslide case unit: (a) ANN; (b) RF.

Figure 13. Summary plot of SHAP values: (a) ANN; (b) RF.

Ngo PTT, Panahi M, Khosravi K, Ghorbanzadeh O, Kariminejad N, Cerda A, Lee S. 2021. Evaluation of deep learning algorithms for national scale landslide susceptibility mapping of Iran. Geosci Front. 12(2):505–519.

 Web of Science ®Google Scholar

Wang Y, Fang ZC, Hong HY. 2019a. Comparison of convolutional neural networks for landslide susceptibility mapping in Yanshan County, China. Sci Total Environ. 666:975–993.

 PubMed Web of Science ®Google Scholar

Hong HY, Shahabi H, Shirzadi A, Chen W, Chapi K, Bin Ahmad B, Roodposhti MS, Hesar AY, Tian YY, Bui DT. 2019. Landslide susceptibility assessment at the Wuning area, China: a comparison between multi-criteria decision making, bivariate statistical and machine learning methods. Nat Hazards. 96(1):173–212.

 Web of Science ®Google Scholar

Wang Y, Sun DL, Wen HJ, Zhang H, Zhang FT. 2020. Comparison of random forest model and frequency ratio model for landslide susceptibility mapping (LSM) in Yunyang County (Chongqing, China). IJERPH. 17(12):4206.

 Web of Science ®Google Scholar

Chen W, Chen X, Peng JB, Panahi M, Lee S. 2021. Landslide susceptibility modeling based on ANFIS with teaching-learning-based optimization and Satin bowerbird optimizer. Geosci Front. 12(1):93–107.

 Web of Science ®Google Scholar

Moosavi V, Niazi Y. 2016. Development of hybrid wavelet packet-statistical models (WP-SM) for landslide susceptibility mapping. Landslides. 13(1):97–114.

 Web of Science ®Google Scholar

Sun DL, Xu JH, Wen HJ, Wang DZ. 2021. Assessment of landslide susceptibility mapping based on Bayesian hyperparameter optimization: a comparison between logistic regression and random forest. Eng Geol. 281:105972.

 Web of Science ®Google Scholar

Sun D, Wen H, Wang D, Xu J. 2020a. A random forest model of landslide susceptibility mapping based on hyperparameter optimization using Bayes algorithm. Geomorphology. 362:107201.

 Web of Science ®Google Scholar

Sun DL, Xu JH, Wen HJ, Wang Y. 2020c. An optimized random forest model and its generalization ability in landslide susceptibility mapping: application in two areas of Three Gorges Reservoir, China. J Earth Sci. 31(6):1068–1086.

 Web of Science ®Google Scholar

Pham BT, Trung NT, Qi CC, Phong TV, Dou J, Ho LS, Le HV, Prakash I. 2020. Coupling RBF neural network with ensemble learning techniques for landslide susceptibility mapping. Catena. 195:104805.

 Web of Science ®Google Scholar

Ali SA, Parvin F, Vojtekova J, Costache R, Linh NTT, Pham QB, Vojtek M, Gigovic L, Ahmad A, Ghorbani MA. 2021. GIS-based landslide susceptibility modeling: a comparison between fuzzy multi-criteria and machine learning algorithms. Geosci Front. 12(2):857–876.

 Web of Science ®Google Scholar

Zhou X, Wen H, Zhang Y, Xu J, Zhang W. 2021. Landslide susceptibility mapping using hybrid random forest with GeoDetector and RFE for factor optimization. Geosci Front. 12(5):101211.

 Web of Science ®Google Scholar

Yi YN, Zhang ZJ, Zhang WC, Jia HH, Zhang JQ. 2020. Landslide susceptibility mapping using multiscale sampling strategy and convolutional neural network: a case study in Jiuzhaigou region. Catena. 195:104851.

 Web of Science ®Google Scholar

Wang Y, Wen H, Sun D, Li Y. 2021. Quantitative assessment of landslide risk based on susceptibility mapping using random forest and GeoDetector. Remote Sens. 13(13):2625.

 Google Scholar

Kavzoglu T, Sahin EK, Colkesen I. 2015. Selecting optimal conditioning factors in shallow translational landslide susceptibility mapping using genetic algorithm. Eng Geol. 192:101–112.

 Web of Science ®Google Scholar

Sun DL, Gu QY, Wen HJ, Shi SX, Mi CL, Zhang FT. 2022b. A hybrid landslide warning model coupling susceptibility zoning and precipitation. Forests. 13(6):827.

 Web of Science ®Google Scholar

Sun D, Gu Q, Wen H, Xu J, Zhang Y, Shi S, Xue M, Zhou X. 2022a. Assessment of landslide susceptibility along mountain highways based on different machine learning algorithms and mapping units by hybrid factors screening and sample optimization.

 Google Scholar

Data availability statement

The data presented in this study can be available on request from the corresponding author.