Fig 1.
Geographic map of Vietnam (right) marked by the study area overlying the regional topography.
Left: study area sub-regions, corresponding to the areas impacted by Typhoon Ketsana and Molave, and cyclone Podul, contain the color-coded landslide inventories used in the investigation. Source: Vietnam administrate boundaries are available in the Humanitarian Data Exchange website. DOI: 10.6084/m9.figshare.25710555.
Table 1.
Summary of the satellite imagery used for landslide mapping.
Table 2.
Summary of variables used in the development of landslide susceptibility models.
Fig 2.
Variables (geo-factors) used in developing the Random Forest models.
TWI and TPI are the Topographic Wetness and Position Indices, respectively. Rocky types (lithology) across the study are are differentiated as 1 = Mafic Metamorphic Rocks with quartz-poor component, 2 = Igneous—Intrusive Acid-to-Neutral Rocks, 3 = Igneous—Intrusive Mafic-to-Ultramafic Rocks, 4 = Felsic Metamorphic Rocks with quartz-rich component, 5 = Igneous- Extrusive Mafic-to-Ultramafic Rocks, 6 = Sedimentary Clastic Rocks, and 7 = Quaternary Sediments. Data source: The 30-meter NASADEM can be accessed from Earth Data. Elevation, Slope, Aspect, TWI, TPI, and Drainage density were derived from the NASADEM. Rainfall data was accessed from the website of Climate Hazards Group InfraRed Precipitation with Station data CHIRPS. The lithology and distance to fault data are included in the supplementary material (S2 Data). DOI: 10.6084/m9.figshare.25710138.
Table 3.
Reclassified lithological units in the study area.
Fig 3.
Landslide susceptibility maps of the study area were developed using the RF algorithm for (a) post-Ketsana, (b) post-Podul, and (c) post-Molave periods.
A gradual increase in landslide susceptibility is evident after each typhoon/storm event in the past decades. Data Source: District boundary, https://cmr.earthdata.nasa.gov, https://www.chc.ucsb.edu/data/chirps, Institute of Geological Sciences, (VAST). DOI: 10.6084/m9.figshare.25710573.
Fig 4.
Maps illustrating the absolute change in probabilistic landslide susceptibility values following the impact of (a) Storm Podul and (b) Typhoon Molave in comparison to the post-Ketsana susceptibility conditions.
Positive values denote an increase in landslide susceptibility, while negative values indicate a decrease in susceptibility compared to the post-Ketsana condition. District boundary, https://cmr.earthdata.nasa.gov, https://www.chc.ucsb.edu/data/chirps, Institute of Geological Sciences, (VAST). DOI: 10.6084/m9.figshare.25710582.
Fig 5.
Plots showing Mean Decrease in Accuracy of the three landslide susceptibility models developed using the RF algorithm for the (a) post-Ketsana, (b) post-Podul, and (c) post-Molave periods.
The plots show the relative importance of different variables in predicting landslide susceptibility in each period.
Fig 6.
Partial Dependence Plots illustrate variability in predicted landslide probability resulting from changes in the values of individual variables.
Table 4.
Out-of-bag (OBB) error rates of the RF models and other performance evaluation metrics as calculated for the test data.
Fig 7.
The AUC shows the plot of true positive and false positive percentages of the test data for different RF models.
The AUC values of 96.6%, 95.7%, and 96.7% yielded for the post-Ketsana, post-Podul, and post-Molave landslide susceptibility models, respectively.
Fig 8.
Mean prediction pixel values of RF plotted against the standard deviation of mean pixel values for all individual pixels for the post-Molave landslide susceptibility model.
Distribution pattern of mean and standard deviation of the predictions remain same for all landslide susceptibility models i.e., they exhibit a quadratic relationship.
Fig 9.
Maps depicting the uncertainty of RF predictions at the pixel level for (a) post-Ketsana, (b) post-Podul, and (c) post-Molave landslide susceptibility models.
The uncertainty is estimated by computing the standard deviation of predictions from 500 trees for each pixel across the entire study area. District boundary, https://cmr.earthdata.nasa.gov, https://www.chc.ucsb.edu/data/chirps, Institute of Geological Sciences, (VAST). DOI: 10.6084/m9.figshare.25710606.