**6. Discussion**

This research was proposed to apply the TFN-AHP method to map landslide susceptibility in Kenya. Figure 5 directly displays the visible landslide susceptibility information for the entire Kenyan territory, indicating the likelihood of potential landslides. As a developing country, such information would greatly benefit Kenyan e fforts to minimize landslide-induced losses and develop optimized land managemen<sup>t</sup> policies. From Figure 6, it was observed that 44.6% of Kenya is classified as highand extremely high-susceptibility zones, whereas 26.11% of Kenya was mapped as having low and very low susceptibility. High and extremely high landslide susceptibility zones predominantly cover the rift valley region and its surrounding areas. This finding can be attributed to plentiful rainfall, steep terrains, and fractured ground. Low and very low landslide susceptibility areas are primarily distributed in the southwestern and coastal regions. The distribution of susceptibility classifications also varies in di fferent provinces (Figure 7). The Rift Valley Province had a majority of the historical landslides. This province has the largest area coverage of extremely high- and high-susceptibility zones.

Dozens of methods have been used in LSM at di fferent scales. For large areas with poor availability of historical landslide inventories, the spatial multicriteria evaluation (SMCE) method has exhibited overwhelming advantages over statistical and the physically based methods [15,42]. As a representative SMCE method, a review of previous studies (as displayed in Table 1) has suggested that the AHP and its fuzzy extensions are one of the favorable methods in LSM for large areas (e.g., for a whole country). One limitation of the application of the statistical method in this study is the incompleteness of the historical landslide inventory, which reduces the reliability of the results. Despite this, the historical landslide inventory can still be used for validation in a better than no sense.

The validation results demonstrate that the adopted TFN-AHP resulted in promising accuracy with an AUC value of 0.86 (Figure 9). Despite no strict rules for the evaluation of this accuracy, the resultant accuracy seems to be good compared to similar studies in di fferent areas [1,3,8]. For the LSM, an ideal result map should include as many historical landslides as possible in "high" or "extremely high" susceptibility regions. Additionally, few historical landslides should occur in the "low" or "very low" susceptibility region. Figure 8 shows that the concentration of known landslides decreases from the extremely high category to the very low category. For decision making under multiple criteria, it is di fficult for humans to quantify criteria weights using extract numbers. However, rational decisions can be made by skilled experts through a certain value with some uncertainties to

capture human subjectivity. Given this, the TFN-AHP makes the comparison process more flexible to minimize the objectivities and uncertainties involved in the conventional AHP process. For the purpose of comparison, the map produced using the conventional AHP is illustrated in Figure 10. The ROC analysis, with an AUC value of 0.72, is also plotted in Figure 9. To perform the conventional AHP, the element value of the TFN-AHP comparison matrix is replaced by a single number according to Table 2. The comparison matrix for the conventional AHP analysis is shown in Appendix A. It should be noted that another source of the subjectivities involved in this study and other similar studies using the SMCE methods or statistical methods may originate from the selection of the LCFs. As shown in Table 1, the number of factors used for LSM ranges from 3 to 10. As discussed in many case studies [1,5,9,16,25,27] and more recently reviewed in [13], the selection of LCFs largely depends on conditions such as data availability, scale, and nature of the study area.

**Figure 10.** The landslide susceptibility map produced using the conventional AHP.

Even for skilled departments from China and many other developing countries, no universal rules have been proposed. Hence, for a given study area, comparative studies have always been conducted to select the best maps.
