Enhancing rockfall modelling through an integrated workflow, from source area definition to susceptibility zoning

The main complexity of rockfall modelling lies in the need for a series of dedicated methodological choices and assumptions. Despite specific aspects of modelling have been largely discussed in the literature, a comprehensive methodology to assess susceptibility posed by rockfalls is still missing. To fill this gap, we have proposed a novel workflow in this study, including methods for identifying source areas, deterministic runout modelling, classifying runout modelling output to establish an objective rockfall probabilistic susceptibility zonation, and comparing and validating the results. This methodology is applied to the island of El Hierro (Canary Islands, Spain), where rockfalls pose a significant threat to structures, infrastructure, and the population.

In the first stage, three different approaches were proposed to identify rockfall source areas, ranging from scenarios with limited data availability to those with extensive topographic, geological, and geomorphological information. The first approach employed a morphometric criterion, establishing a slope angle threshold to identify source areas. The second approach used a statistical method employing Empirical Cumulative Distribution Functions (ECDF) of slope angle values. The third method employed a probabilistic modelling framework that combined multiple multivariate statistical classification models, using mapped source areas as dependent variables and thematic information as independent variables.

Subsequently, a rockfall simulation was carried out using a physically based model using the maps of the three source areas as input. A key result of the rockfall modelling simulations was the rockfall trajectory count maps. These maps, highlighting areas prone to rockfall on El Hierro, indicated the probability that a given pixel would be affected by these processes.

Then, this study also explores the strategies to validate the rockfall susceptibility model outputs, using different types of inventories. Therefore, to get susceptibility maps with a probabilistic approach, two classification methods were applied: unsupervised and supervised statistical techniques using distribution functions. The unsupervised classification used only the raster map of rockfall trajectory counts, while the supervised classification considered additional data on areas already affected by rockfalls.

Diffused metrics comparing modelled and observed values (i.e., ROC plots and correspondent AUCROC) can be used to show the performances of susceptibility models, regardless the adopted classification approach. Finally, the six susceptibility maps were compared to emphasize the impact of source area definition on the distribution of rockfall trajectories.

In summary, the methodology proposed provides guidance for an objective and reliable rockfall modelling, supporting civil protection, emergency authorities, and decision-makers in evaluating and assessing potential rockfall impacts. This contributes to enhanced rockfall hazard assessments and improved mitigation strategies on the island of El Hierro and potentially in similar geological settings globally.