Assessment of landslide susceptibility triggered by precipitation in the Metropolitan Region of Recife, Brazil

Abstract

Landslide events can severely impact communities, underscoring the need to understand local dynamics and rainfall-related triggers when assessing susceptibility. This study pursued three main objectives: (1) to identify critical precipitation thresholds for landslide-triggering rainfall in the Recife Metropolitan Region (RMR), Brazil, from 2016 to 2022, using two empirical methods— the event duration (ED), based on accumulated precipitation, and the intensity–duration (ID), based on rainfall intensity over specific periods; (2) to create a comprehensive landslide inventory map; and (3) to develop a regional susceptibility map using machine-learning algorithms (RF, KNN, XGBoost, and NB) to identify key destabilizing factors. A total of 221 landslides were mapped across the region. The ED method showed stronger correlations with local landslide occurrences (R² > 97.5 %) compared to the ID method (R² < 90.5 %). Susceptibility modeling incorporated nine factors, with land use, elevation, and geomorphology emerging as the most influential in slope instability. The resulting susceptibility map was classified into five categories: very low (28 %), low (35 %), moderate (27 %), high (9 %), and very high (1 %). Model performance was validated using ROC analysis, with the RF model achieving excellent predictive capability (AUC = 0.905). These findings offer valuable insights for landslide risk management in the study area, highlighting the importance of accumulated precipitation and terrain characteristics. The susceptibility map offers an evidence-based foundation for proactive measures to mitigate landslide impacts.