Landslide inventory of the 2023 Serra do Mar disaster (Brazil)

Abstract

Abstract Landslides pose severe hazard to populations and infrastructure in Brazil, with a notable recurrence in the South and Southeast coast, characterized by mountainous reliefs and high annual rainfall. In February 2023, the north coast of the São Paulo State experienced an extreme rainfall event that triggered hundreds of landslides, resulting in 46 deaths and thousands of affected. This article objective was to prepare a landslide inventory of this event for the two areas more intensely affected by landslides. We evaluated the morphological and land use characteristics of the affected areas, as well as the landslide susceptibility depicted in official maps and population density of affected areas. Landslide mapping was performed with an object-based Random Forest algorithm. The inventory achieved high accuracy (0.871 for sensitivity; 0.986 specificity) and a total area of 1.91 km2 was affected by landslides, with 496 initiation points identified. Most initiation points (238) are located in areas previously mapped as highly susceptible, associated with forest formations. Landslides were triggered mainly on slopes with angle from 27º to 30º and topographic wetness index between 3.5 and 5.5. Initiation points occurred in areas with less than 100 people/km2, but runout and deposition zones reached densely populated areas (> 1000 people/km2). We conclude that, despite most landslides were initiated in forested areas, correctly identified in previous maps as highly susceptible, the runout zone reached lower susceptibility areas. This finding highlights the necessity for spatially-explicit hazard maps in Brazil that indicate the potential landslide trajectory, as a tool to support efficient risk mitigation policies.

Keywords

• Southeast Brazil
• Random forest
• OBIA
• Morphometry
• Exposed population