Identification of rockfall source areas using the seed cell concept and bivariate susceptibility modelling

Abstract

The objective of this research was to prepare a rockfall susceptibility map. Explorations were conducted in the Dubračina River basin (Croatia). The input data included a geological map, an orthophoto and a 1-m digital terrain model (DTM). After a talus inventory was prepared, the seed cell concept was applied to define the rockfall source areas. The contributing factors (predictors) of rockfalls were evaluated by the chi-squared test. The analysis confirmed the following predictors: CORINE land cover, lithology, slope, aspect, distance from a spring, distance from a road, distance from a fault, distance from a stream, and distance from the rock-soil geological boundary. A matrix pairwise comparison of the predictor ratings was used to define the most significant contributing factors. The predictors that affected the susceptibility map in the share of 86.3% were the slope (61.6%), lithology (13.4%), CORINE land cover (6.2%), and distance from the rock-soil geological boundary (5.1%). Two susceptibility maps were prepared: one using all nine contributing factors and another using the four most significant factors. The analysis showed that both maps were good, with the same areas under the receiver operating characteristic (ROC) curves. The map prepared with only four contributing factors can be considered a better map due to its more precise spatial definition of critical areas. It can be concluded that geological map, 1-m DTM and orthophoto provide enough data to prepare reliable rockfall susceptibility map. The application of the bivariate statistical zonation method called the “frequency ratio method” was proven to be successful. This research demonstrates that the application of the seed cell concept can be useful to speed up the process of rockfall source area detections in large research regions.

Rockfall Susceptibility Assessment and Landscape Evolution of San Nicola Island (Tremiti Islands, Southern Adriatic Sea, Italy)

“…Machine learning techniques, finally, are becoming more popular given their ability to deal with large datasets and derive complex non-linear patterns [26]. In the specific case of rockfalls, research has shown rapid growth in recent years, with a clear increasing trend of annual publications on the topic, and several works based on different datasets and techniques have recently been published [26][27][28][29][30][31][32].…”