*Model Development and Validation*

The landslide inventory data was randomly categorized into two datasets, i.e., training (70%) and testing (30%). The landslide susceptibility map based on 8 causative factors using AHP is depicted in Figure 10. The map was divided into 4 classes: very low (0–0.24), low (0.25–0.49), moderate (0.5–0.74) and high (0.75–1) according to natural breaks to define the class intervals in the susceptibility map. From total of 242 landslides, more than 78.1% falls under the high zone. Whereas 11.98% falls under the moderate zone and combined 9.92% falls under the low and very low zones (Table 5).

**Figure 10.** Landslide Susceptibility Map of the Study Region.

**Table 5.** Landslide Susceptibility Results for the Samdrup Jongkhar–Trashigang Highway Region.


Validation of the susceptibility maps was performed for randomly selected data from the inventory data using receiver operating characteristics (ROC). This is an effective way to analyze the quality of predictive techniques [81]. The ROC curve is plotted between true positive rate (sensitivity) on the Y

axis against false positive rate (specificity) on the X axis. The terms "sensitivity" and "specificity", which are used to plot ROC curves, are defined as follows.

$$\text{Sensitivity} = \frac{\text{TP}}{\text{TP} + \text{FN}} \tag{7}$$

$$\text{Specificity} = \frac{\text{TN}}{\text{FP} + \text{TN}} \tag{8}$$

where true positive (TP) is the number of actual landslides predicted correctly, and true negative (TN) is the total number of non-occurring landslides predicted correctly. False positive (FP) is the number of actual landslides inaccurately predicted as non-occurring landslides, and false negative (FN) is the number of non-occurring landslides inaccurately predicted as actual landslides. [82]. The area under the ROC curve (AUC) was also used to determine the quality of the prediction by analyzing the model's ability to forecast the occurrence or nonoccurrence of predefined events [83]. The results of the success rate curve of the AHP model had an AUC of 0.798, corresponding to a prediction accuracy of 79.8% (Figure 11).

**Figure 11.** Receiver operating characteristic (ROC) curve of the susceptibility map by AHP.

The results of pairwise comparison, priority estimation, and ranking of all the criteria could be applied to other study areas for susceptibility assessment. Several high-resolution satellite image datasets are required to better understand the locations and perform these assessments. In cases of unavailability of high-resolution satellite images, Google Earth images could be useful for preparation of some indicators. The use of Google Earth images could also be helpful for accurate identification of landslide locations and conduction of extensive field studies. All the requirements for more accurate analysis depend on study location tectonics, data availability, and proper methods. Therefore, based on local geology and tectonic conditions, these results could be transferable and applicable in other locations in both small- and large-scale areas.

### **6. Conclusions**

Landslides are the most frequently occurring natural hazards, especially in the Himalayan regions, which suffer from heavy monsoonal rainfall and subsequent landslides. In this study, the temporal probability of landslide events was determined using rainfall and landslide data from 2014–2017 along Samdrup Jongkhar–Trashigang highway in East Bhutan. The highway was divided into three zones based on land use, topography, and rain gauge coverage for the determination of temporal probability. Thereafter, a landslide susceptibility map was developed using AHP. The results of temporal probability

were validated with landslide event dataset of 2018 to understand the applicability of the model. The conclusions from the study can be summarized as follows.


The present study on rainfall threshold estimation and the development of a susceptibility map for eastern Bhutan along Samdrup Jongkhar–Trashigang highway is an important study in the context of the Bhutan Himalayas, for which study on both these aspects is lacking. The current study can be regarded as a preliminary step towards risk management, which could be supported by conducting a hazard and vulnerability assessment of the region. The temporal probabilities determined can be integrated with the susceptibility map to obtain a landslide hazard map. However, the results might be improved by increasing the number of landslide events and using precipitation data with a higher temporal resolution. The results from the present study also could prove helpful to civic authorities in identifying key sections of the road which are most vulnerable to landslides, and undertaking strict measures to prevent slope failures, strengthening the transportation network and saving human lives.

**Author Contributions:** Conceptualization, A.D. and R.S.; methodology, A.D. and R.J.; data curation, D.D.; writing—original draft preparation, A.D. and R.S.; writing—review and editing, B.P. and A.M.A.; supervision, B.P. and R.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the Centre for Advanced Modelling and Geospatial Information Systems (CAMGIS) in the University of Technology Sydney (UTS) under Grants 321740.2232335 and 321740.2232357; Grant 321740.2232424, and Grant 321740.2232452. This research is also supported by Researchers Supporting Project number RSP-2019/14, King Saud University, Riyadh, Saudi Arabia. Also, the research was funded from BRACE project (NERC/GCRF NE/P016219/1).

**Acknowledgments:** The authors are thankful to National Center for Hydrology and Meteorology, Royal Government of Bhutan for providing rainfall data and the Border Roads Organization (Project DANTAK), Government of India for providing landslide data. Authors are also thankful to staff of College of Science and Technology, Royal University of Bhutan, who helped directly or indirectly while carrying out the present study. We also thank the editor and the reviewers for reviewing and suggesting valuable modifications, which has helped in improving the quality of this paper.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
