Table 1.
Comparative perspective of studies integrating landslide susceptibility and social vulnerability.
Table 2.
General characteristics of landslide susceptibility models.
Table 3.
Common copula families and their bivariate properties.
Table 4.
Spatial and social input variables used in the landslide susceptibility analysis.
Fig 1.
Inputs used in the estimation of the landslide susceptibility map.
Inputs include DEM, aspect, roughness, land cover, tau DEM and landslide mask.
Fig 2.
The diagram show how inputs are transformed into landslide susceptibility maps through several algorithms, evaluated, and subsequently associated with socioeconomic variables.
Fig 3.
Correlation matrix for every causative factor.
Table 5.
Model performance based on AUC values and sensitivity to multicollinearity.
Fig 4.
AUC and ROC plots for every tested model.
This matrix plot shows every AUC and ROC for easy comparison purposes.
Fig 5.
Landslide susceptibility maps created by each tested model.
Matrix plot allows for quick comparison between susceptibility maps.
Fig 6.
Raw susceptibility map generated with XGBoost.
Full high-resolution susceptibility map of the champion model.
Fig 7.
Uncertainty quantification plots.
Maximum and minimum plots per areal unit as uncertainty quantification method, to portray the whole range of values that may arise.
Fig 8.
Categorized susceptibility map.
Susceptibility values categorized to create a simpler-to-read map.
Table 6.
Classification of landslide susceptibility values into categorical classes.
Fig 9.
CENAPRED official susceptibility map.
Notice that CENAPRED categorization values do not necessarily coincide with our methodology, since we intend to highlight different zones. This implies that the range and scope of our categorical map may differ from that of CENAPRED, yet this map is presented for transparency purposes.
Fig 10.
Feature importance, Importance gain, and Geodetector ranks features according to their importance.
Fig 11.
Combined landslide susceptibility plots.
Susceptibility tends to increase to the west, therefore, as it increases, the corresponding AGEBs are located to the west.
Fig 12.
SLI vs landslide susceptibility.
Social lag index plotted against different landslide susceptibility values.
Fig 13.
Beeswarm plot for the entire dataset.
Table 7.
Copula model performance based on log-likelihood and association.
Table 8.
Segment-wise model performance and SHAP-based importance of the Social Lag Index (SLI).
Fig 14.
SHAP plots for changepoint-induced partitions.
Beeswarm plot using changepoint induced partitioning.
Table 9.
Segment-wise copula model selection based on log-likelihood.
Table 10.
R2 and permutation importance by decile.
Fig 15.
SHAP plots for decile-induced partitions.
Beeswarm plot for each of the 10 partitions generated.
Table 11.
Decile-wise copula model selection based on log-likelihood and dependence.
Fig 16.
Boxplots portray first quartile, median, third quartile, and outlier data.
Fig 17.
Barplots shows the mean for each considered variable as separated by susceptibility class.
Table 12.
Feature importances from XGBoost model.
Fig 18.
SHAP plots for each variable that was used to construct the Social Lag Index.
Beeswarm plots showing potential association for individual social variables.
Table 13.
Copula fit per variable (variable IDs from master list).