Slope instability research and susceptibility mapping is a fundamental component of hazard management and an important
basis for provision of measures aimed at decreasing the risk of living with landslides. On this basis, this paper presents the result of
a comprehensive study on slope stability analyses and landslide susceptibility mapping carried out in part of Sado Island of Japan.
Various types of landslides occurred in the island throughout history. Little is known about the triggering factors and severity of
old landslides, but for many of the recent slope failures, the slope characteristics and stratigraphy are such that ground surfaces
retain water perennially and landslides occur when additional moisture is induced during rainfall and snowmelt. A range of
methods are available in literature for preparation of landslide susceptibility maps. In this study we used two methods namely, the
analytical hierarchy process (AHP) and logistic regression, to produce and later compare two susceptibility maps. AHP is a semiqualitative
method, which involves a matrix-based pair-wise comparison of the contribution of different factors for landsliding.
Logistic regression on the other hand promotes a multivariate statistical analysis with an objective to find the best-fitting model
that describes the relationship between the presence or absence of landslides (dependent variable) and a set of causal factors
(independent parameters). Elevation, lithology and slope gradient were casual factors in this study. The determinations of factor
weights by AHP and logistic regression were preceded by the calculation of class weights (landslide densities) based on bivariate
statistical analyses (BSA). The differences between the AHP derived susceptibility map and the logistic regression counterpart are
relatively minor when broad-based classifications are considered. However, with an increase in the number of susceptibility
classes, the logistic regression map gave more details but the one derived by AHP failed to do so. The reason is that the majority
of pixels in the AHP map have high values, and an increase in the number of classes gives little change in the spatial distribution
of susceptibility zones in the middle. To verify the practicality of the two susceptibility maps, both of them were compared with a
landslide activity map containing 18 active landslide zones. The outcome was that the active landslide zones do not completely fit
into the very high susceptibility class of both maps for various reasons. But 70% of these landslide zones fall into the high and
very high susceptibility zones of the AHP map while this is 63% in the case of logistic regression. This indicates that despite the