**1. Introduction**

Every year, landslides cause a large number of deaths and enormous property losses in mountainous areas [1]. Landslides are a prehistoric issue. It is currently receiving considerable attention since damage induced by landslides has risen in recent years. Estimated fatalities from landslides reached 32,322 between 2004 and 2010, though this value is likely underestimated [2]. The situation varies in different countries. Landslides are concentrated in developing countries or regions, such as the Himalayan region and its surrounding areas in China and African countries. Hence, more effort is required to reduce landslide risks within those countries. Within this topic, the preemptive identification of landslide-prone areas through landslide susceptibility mapping (LSM) is a very promising hazard mitigation approach.

> 7

The term landslide susceptibility is a quantitative measure of the likelihood of slope failures under a particular geological condition [3]. With the increasing availability of geospatial data and rapid developments in computational science, numerous LSM methods have been proposed in the last three decades. Most of these methods were built on geographic information systems (GISs). In a broad sense, these LSM models can be summarized as qualitative (knowledge-based or inventory-based) and quantitative (statistically or physically based). In qualitative LSM modeling, each landslide factor is weighted based on the knowledge of experts in geotechnical or geological fields. Afterward, the derived weights were combined to calculate the landslide susceptibility index (LSI). Typical qualitative LSM models include heuristic analysis, inventory analysis, and analytic hierarchy processing (AHP) [4]. As a comparison, statistical LSM models quantify the weights of each factor based on the spatial correlations of historical landslides and these factors. Building on the basic assumption that "the past predicts the future", the weights determined using historical landslides are used to predict the likelihood of future landslide occurring. Frequency ratios [5], logistic regressions [6], weights of evidence [7], artificial neural networks [8], and support vector machines [9] are frequently used statistically based LSMs in the literature. For physically based LSM models, slope stability models and groundwater flow models are integrated to calculate the safety factor for each slope unit. Several programs have been developed for LSM, such as SHALSTAB, SINMAP, and TRIGRS [10,11]. The advantages and disadvantages of different LSM models have been reviewed by Van Western et al. [12] and Reichenbach et al. [13]. Comparative studies have shown that the optimized selection of LSM methods largely depends on the scale, nature, and data availability of the study area [14–16].

Similar to most African countries, landslide is ranked as the deadliest geohazard in Kenya [17,18]. Despite enormous damage induced by landslides, literature reviews indicate that very few attempts have been carried out to research landslides in Kenya. The studies of landslides performed in Kenya in the past few decades have concentrated on landslide inventory mapping [19], geological investigation of single landslide events and developing general overviews of landside phenomena [18]. It is noted in the literature that intensive precipitation is a dominant factor triggering landslides in Kenya [20]. Steep topography, weathered regolith, and human activities such as deforestation, overgrazing, and overfarming have been identified as causative factors of landslides in Kenya [18]. In recent years, the continuously growing population and expansive development of infrastructure have placed a heavy burden on the environment and land resources in Kenya. No systematic research on landslide susceptibility assessment in Kenya has been published yet. Filling this research gap is the reason why this study was performed.

Difficulties remain in developing LSM for the whole territory of a country because of inadequate availability of landslide inventories and related information. As illustrated in Table 1, a literature review of some such examples showed that qualitative methods, such as spatial multicriteria evaluation (SMCE) and heuristic weighting, are the most popular existing LSM on a national scale. Van Western et al. [10] suggested that the most suitable methods for LSM at a medium scale are quantitative methods, while qualitative methods are more appropriate for LSM of large areas (small scale) [19]. The cell size of LSMs varies from coarse (1000 m) to medium (30 m) for qualitative and quantitative LSMs. The suitability of cell size is typically determined by data availability and the mapping scale [20].

The main objective of this work was to develop a landslide susceptibility map for Kenya. To conduct this, the fuzzy analytic hierarchy process (FAHP) method was adopted. This FAHP is a semiqualitative method suitable for LSM on a national scale. The landslide inventory and landslide causative factors used in this study were collected from a variety of sources. Regions highly susceptible to landslides in Kenya were highlighted as a basis for further studies of landslide hazards or risk assessments. Additionally, the output presented serves as an effective tool for the authorities involved in land planning and land resource management.


**Table 1.** Studies of landslide susceptibility mapping at nation scale.
