Landslide susceptibility assessment in Limbe (SW Cameroon): A field calibrated seed cell and information value method

The dissected volcanic terrains around Limbe, SW Cameroon are frequently affected by small scale but destructive landslides. In this study, a raster-based data driven method involving seed cells is used to build a landslide susceptibility model for the Limbe area. Factors considered to be potential controls of slope failure within this area include slope gradient, rock type, distance from roads, slope orientation, mean annual precipitation, soil type, land cover type, stream density and distance from stream. 63 small to very small translational and rotational landslide scars were identified through extensive field work. Landslide data is randomly divided into a training (75%) and validation set (25%) and seed cells are generated by creating 25m buffer zones around the head scarp of each scar. The quantitative relationship between landslide seed cells and the above-mentioned factors is established by a data driven approach to obtain weighted factor classes. Summing weighted factor layers, a continuous scale of susceptibility indices is obtained and reclassified into 5 susceptibility classes. Seed cells obtained from the validation data set were used to evaluate the quality of several models involving different controlling factors. Our preferred model combines the weight of 6 factors (i.e. slope gradient, land cover, mean annual precipitation, stream density, proximity to roads and slope orientation). 78% of the validation seed cells are located within the high to very high susceptibility class, which occupy 16.9% of the study area. The obtained susceptibility map is combined with the outline of urban areas and key infrastructures to evaluate zones that are vulnerable to the impact of future slope failures. Such an approach will assist civil protection and urban planning efforts in SW Cameroon. ► We used a data driven bivariate statistical model to construct a landslide susceptibility model for the Limbe area. ► Model has a high predictive power. ► Model is advantageous because it is reproducible and limits expert subjectivity. ► Susceptibility map enables to identify settlements and elements at risk.