**5. Results**

#### *5.1. Weights of LCFs and Their Subclasses*

The weighting vector derivation plays a central role in multicriteria decision making. For the present study, weights were assigned to each LCF. For this purpose, the geotechnical experts were called upon to make a pairwise comparison of the LCFs based on their experiences and knowledge. As illustrated in Table 4, the pairwise comparison matrix for the ten LCFs was constructed by considering expert opinion and similar previous studies [21–23,26,27]. From the matrix, the weight vector for the criteria was computed using Equation (14) to Equation (17) and is presented in Table 5. After normalization, the weights for each criterion were derived using Equation (18) and are shown in Table 6. The *CR* was calculated using Equations (3) and (4). When the *CR* = 0.086 < 0.1, the judgment was deemed to be consistent.

For the subcriteria (subclasses) under a certain uplevel criterion (an LCF), the weights were derived using the same procedures. The sum of subcriteria weights under each corresponding uplevel criterion should be 1.0. Hence, 10 comparison matrices were created. Additionally, the final weights for the subclasses within each LCF were calculated and are shown in Table 6. Before using these calculated weights, a consistency check was conducted for each comparison matrix. Only if CR < 0.1 was the derived weight accepted. Since all CR values were less than 0.1 (Table 6), a consistency check of the 10 matrices indicated that all the judgments were consistent.

From the TFN-AHP analysis, slope gradient (0.1923), MAP (0.1884), and curvature (0.1651) were considered to be the three most important factors contributing to landslide occurrence, whereas the least important factors were land use (0.0220) and aspect (0.0315). For the factor of slope gradient, the terrain steeper than 30◦ was most susceptible to landslides (weight for this subclass is 0.364313), while the category of 5–10◦ obtained the lowest weights (0.071825) in determining the landslide occurrences. It also can be seen from results that barren land (0.160743), bush land (0.136464), and grassland (0.121685) were most susceptible to landslides compared with other land use types. In case of curvature, both concave and convex terrain were more prone to landsliding than flat area. Convex terrain (subclass weight is 0.570014) was more favorable for landsliding than concave terrains (0.356956).


*S*Landform 0.9008 1.0000 0.9288 1.2194 1.0000 1.0000 1.0000 1.0000 – 1.0000 *S*Slope 0.1146 0.9796 0.1637 0.4773 0.4723 0.3505 0.8585 0.5674 0.2155 – min{*V*(*Si* ≥ *Sj*)} 0.1146 0.9796 0.1637 0.4773 0.4723 0.3505 0.8585 0.5674 0.2155 1.0000


#### *5.2. Landslide Susceptibility Maps*

Using Equation (18), the LSI value for each raster cell within Kenya was calculated. As shown in Figure 5, the resultant LSI map was reclassified into five susceptibility levels using the "natural break" function ArcGIS. In total, 15.44% and 29.16% of the Kenyan territory were mapped as extremely high susceptibility zones. A total of 29.16% of the total area was predicted as a high susceptibility zone. Low and very low susceptibility classes covered 20.58% and 5.53% of the study area, respectively. The remaining 29.29% of the study area was determined to be moderately susceptible to landslides (Figure 6). The distribution of susceptibility classes differed in each province. As illustrated in Figure 7, the Rift Valley Province and Eastern Province had the highest percentages of EH landslide susceptibility coverage (21% and 19%, respectively), while the Central Province and Nyanza Province had the lowest percentages of EH landslide susceptibility (5% and 6%, respectively).

**Figure 5.** The landslide susceptibility map produced using the TFN-AHP.

**Figure 6.** Area coverage of the five landslide susceptibility levels in Kenya.

**Figure 7.** Area coverage of the five landslide susceptibility levels in each province.
