Abstract
The Mega Disaster that occurred in May 2024, reaching Rio Grande do Sul (RS), Brazil, resulted in more than 15,000 landslides, affecting nearly 18,000 km2. In this study, we generated a landslide susceptibility map based on historical landslide data (1995-2017) and assessed its capacity to forecast the 2024 landslides. Based on the Random Forest algorithm, we modelled landslide susceptibility for the entire RS with a high overall accuracy of 0.964. The retrospective evaluation using the 2024 landslide inventory considered two sets: i) polygon centroids, and ii) points by pixels. In the first approach, our model correctly classified 69% of the landslides in susceptible areas. The second analysis classified 31% of the points in very high susceptibility areas. Furthermore, when comparing the landslide areas, our model correctly classified most of the landslide rupture zones (79%). Finally, we highlight the need to consider landslide zones for updating susceptibility maps.
Original language | English |
---|---|
Publication status | Accepted/In press - 2025 |
Event | Brazilian Remote Sensing Symposium - Bahia, Salvador, Brazil Duration: 13 Apr 2025 → 16 Apr 2025 Conference number: 21 https://2025.sbsr.com.br/sbsr-2025/page/4706-inicio?lang=en |
Conference
Conference | Brazilian Remote Sensing Symposium |
---|---|
Abbreviated title | SBSR |
Country/Territory | Brazil |
City | Salvador |
Period | 13/04/25 → 16/04/25 |
Internet address |
Austrian Fields of Science 2012
- 105902 Natural hazards
- 102019 Machine learning
- 207402 Remote sensing
Keywords
- natural hazard
- disaster
- Random Forest
- modelling
- Brazil