An object-based approach for semi-automated shallow landslide mapping: suitability and comparison in Itaóca (SP) and Nova Friburgo (RJ), southeastern Brazil

Landslide inventories (geomorphological, seasonal, and event-based) are crucial for susceptibility, vulnerability, and risk assessments. Despite the high frequency of landslides in Brazil, no guidelines, or common procedures for the creation of inventories exist. Object-based image analysis (OBIA) is a well-established method for mapping natural hazards. However, the application of this technique for landslide mapping in Brazil is still incipient. The use of OBIA could improve the creation of landslide inventories in the country. Thus, this study aims to identify rainfall-induced shallow landslides using the OBIA method and compare the parameters used in the classification rule set in two study areas in Brazil: Itaóca (São Paulo state), and Nova Friburgo (Rio de Janeiro state). Both study areas were strongly affected by high-magnitude mass movements in 2014 (Itaóca) and 2011 (Nova Friburgo). The analysis was performed using RapidEye satellite images (5 m resolution), dated 2014/01/30 (Itaóca) and 2011/01/20 (Nova Friburgo), in the eCognition 10.0 (Trimble) software. The classification considered spectral, spatial, and contextual information. The mapping accuracy was assessed by comparison to a shallow landslide inventory created through expert interpretation. The results indicate the good applicability of the OBIA method in the tropical environments of Brazil. In Itaóca, shallow landslides of varied sizes occurred, whereas in Nova Friburgo large landslides were more common, with a medium size of 1.900 m² and 3.200 m² and a median of 725 m² and 950 m², respectively. The rule set applied for both study areas included the same processes but with slightly modified parameters, for example, the mean NDVI and slope were adapted according to the local environmental and geomorphological conditions. The results confirm the transferability of the approach in Brazil, although minor adaptations in the main rule set are required for better results.