Hybrid pixel-based and object-based image analysis approach for landslides rapid mapping: the extreme rainfall in Emilia-Romagna (Italy) May 2023 case study

Landslide rapid mapping inventory, generated in the immediate aftermath of severe landslide events, represents essential information for quantitative landslide risk assessment. It can support effective and efficient response in case of emergency activations, for example by identifying the potential number of affected infrastructures (e.g. roads, railways), and can provide useful information to prioritize the interventions and to guide field survey operations. Additionally, it may represent the base to be used for generating and updating landslide mapping inventories. This research presents a hybrid pixel-based and object-based image analysis approach for landslides rapid mapping using remote sensing optical multispectral imagery. The fully automated supervised procedure relies on change detection analysis, based on quantitative variation of vegetation cover, and can be applied to both aerial imagery and multiple satellites constellations acquisitions at high and very-high spatial resolution. The main novelties rely on the use of an advanced spatial filtering approach, to optimize the removal of commission errors, and a confidence measure, computed using object-oriented image analysis as a solution to optimize the accuracy of the mapping product. Investigated case study for development and testing of the procedure is the heavy rainfall event that hit Emilia-Romagna region (Italy) during May 2023. Satellite imagery acquired by Sentinel-2 and PlanetScope satellites constellations were used to generate rapid mapping products. Validation against two landslides event inventories mapped manually shows increased accuracy of the generated rapid mapping product by applying various threshold values to the confidence measure.