Assessing Temporal Consistency and the Effect of Predisposing Factors in Landslide Susceptibility Models in the Ribeira Quente Valley (São Miguel Island, Azores)

Situated in the North Atlantic Ocean, the Azores is an archipelago of nine volcanic islands, where numerous destructive landslide events have occurred over the past five centuries, triggered by several factors, namely seismic activity, volcanic eruptions, and episodes of intense rainfall. Within this context, this study focuses on the Ribeira Quente valley, located in Povoação Municipality (S. Miguel Island), covering an area of 9,15 km². The valley is highly prone to landslides, which often damage the only road to Ribeira Quente village, leaving it isolated. A major event occurred on October 31^st,1997, when an episode of very intense rainfall triggered nearly 1,000 shallow landslides, primarily translational slides and debris flows. This event resulted in 29 fatalities, the destruction of 36 houses, and left 114 people homeless, while the village became isolated for over 12 hours.

Three historical landslide inventories were developed for this study. The first inventory, based on a 2004 ortophotomap with a resolution of 40 centimeters and a scale of 1:15,000, included approximately 400 landslides. The second inventory, from 2010, was developed using Google Street View, and contained around 250 landslides. Finally, the third inventory, conducted through fieldwork in 2025, identified approximately 260 landslides. In total, the three inventories include around 910 landslides.

Landslide susceptibility analysis provides the essential basis for hazard mapping, a crucial component for quantitative risk assessment. The main objectives of this study are: (i) to investigate whether there is temporal variability in the spatial distribution of landslide susceptibility results; and (ii) to determine the optimal combination of predisposing factors for inclusion in the landslide susceptibility model, maximizing its predictive performance.

Susceptibility modelling was performed using 11 predisposing factors, which were processed as raster datasets with a 5 m × 5 m resolution, alongside historical landslide inventories. To evaluate the influence of each predisposing factor on landslide distribution, factors were hierarchically ranked by their ability to distinguish between terrain units with and without landslides.

The modeling process employed the Information Value method, a bivariate probabilistic approach derived from Bayesian theory. A total of 2,047 susceptibility models were tested for each landslide inventory, and the best model was selected based on its goodness of fit, determined by computing the Success Rate Curves (SRC) and the Area Under the Curve (AUC). The predictive capacity of the best models was then assessed by computing the Prediction Rate Curves and the corresponding AUC.

This study provides essential tools for land-use planning and civil protection. Landslide susceptibility maps can also support the implementation of site-specific risk mitigation measures and prioritize detailed geotechnical investigations. This research is financially supported by the INTERREG program through the PRISMAC project – “Análise, Mitigação e Gestão do Risco de Movimentos de Vertente Potenciados pelas Alterações Climáticas na Macaronésia” (Ref. 1/MAC/2/2.4/0112).