Probabilistic landslide hazard assessments: adaptation of spatial models to large slow-moving earth flows and preliminary evaluation in Loja (Ecuador)

Abstract

Quantitative landslide hazard models provide estimations of the number of landslides per area and time that might be expected in the near future. These models are essential to calculate landslide risk in monetary terms. Although they are very useful tools for managing the activity of unstable slopes, their production calls for a vast amount of spatial and temporal data. Here, we present a case where this was possible producing the quantitative landslide hazard map for the municipality of Loja, Ecuador. It is based on a model that integrates six causal factors (distance to faults, lithology, slope, geomorphology, topographic position index, land use) and a comprehensive multi-temporal inventory of landslides. First, a susceptibility map was generated with a good prediction capability (Area under prediction rate curve, AUPRC: 0.8) combining two widely used and tested probabilistic methods: “Matrix” and “Likelihood ratio”. Subsequently, this map was transformed into a hazard map by including the temporal frequency of landslides. The map assesses the annual probability of each pixel to be set in motion within one of these landslides. The preliminary temporal validation of the hazard map indicates that the pixels mobilized during two years after the map production fit reasonably well with our spatio-temporal forecast. The findings emphasize that classical spatial prediction methods, when augmented by robust and extensive data on landslide distribution and activity, can yield hazard models with reliable predictive capabilities. This suggests that in practical applications, models based on relatively simple calculations can provide effective and reliable starting points for managing landslide risks.