Combining computational models for landslide hazard assessment of Guantánamo province, Cuba

Abstract

As part of the Cuban system for landslide disaster management, a methodology was developed for regional scale landslide hazard assessment, which is a combination of different models. The method was applied in Guantánamo province at 1:100 000 scale. The analysis started with an extensive aerial photointerpretation to produce a landslide inventory map. Five main types of landslide movements were identified: slides (186), rockfalls (22), debrisflows (26), topples (18), and large rockslides (29). The causal factors for each landslide type were analysed using twelve indicators: lithology, geomorphology, landuse, soil, slope angle, internal relief, slope orientation, drainage density, distance to roads and faults, rainfall intensity, and peak ground acceleration. A specific set of causal factors was identified for each of the five landslide types. Artificial Neural Networks (ANNs) were applied for the assessment of susceptibility to slides and the weights of evidence for the assessment of susceptibility to the other landslide types. Success rates were used for evaluating the performance of the model. In all cases more than 80% of landslides were classified within the 10% of most susceptible areas. The five susceptibility maps were reclassified and spatial and temporal probabilities were calculated for each class, based on the landslide densities and on geomorphological evidences. Around 13 % of the province was classified as high susceptible to at least one landslide type, 17 % as moderate and 24 % as low. The results show that it is possible to derive at semiquantitative landslide hazard maps, for different landslide types, making use of a combination of heuristic reasoning and probabilistic derived weights.