AUTOMATIC DETECTION OF LANDSLIDE FEATURES WITH REMOTE SENSING TECHNIQUES: APPLICATION TO MADEIRA ISLAND C. Lira 1 , M. Lousada 1 , A. P. Falcão 2 , A. Gonçalves 2 , S. Heleno 2 , M. Matias 2 , A. J. de Sousa 1 , M. J. Pereira 1 , P. Pina 1 , R. Oliveira 3 and A. B. Almeida 3 IST – Instituto Superior Técnico – Technical University of Lisbon – Lisbon, Portugal CERENA 1 , ICIST 2 , CEHIDRO 3 ABSTRACT The extreme rainfall event of February 20 th , 2010 triggered a series of landslides and alluvium episodes with extensive life and material damages. The use of automatic detection of the landslides over satellite imagery allowed the identification and characterization of the affected areas, the mapping of the landslide features and the calculation of the displaced sediment volume. The study shows that the occurred landslides were shallow, being the basins of Ribeira Brava the most affected areas, with almost two thirds of the total identified landslide polygons. Index Terms— landslides, automatic recognition, remote sensing 1. INTRODUCTION The Madeira Archipelago is located in the north Atlantic Ocean, about 520 km from the African Coast and 1000 km from the European continent. The archipelago includes a serial of small islands around a major one named Madeira, with a total area of 740 km 2 . In terms of geomorphology, Madeira has a very steep topography, with 35% of its territory above 1000 m and 90% above 500 m. Its territory is occupied by a mountain ridge, with very deep "V"-shaped valleys, almost vertical, on harder volcanic rocks strata, extending along the centre region of the island, reaching 1862 m at Pico Ruivo, and the Paul da Serra plateau, a narrow and long planar structure on the western part of the island. The steep characteristic of the slopes (27% to 37% mean slope with areas above 50%) associated to a subtropical climate, with episodes of heavy rain, make the occurrences of landslides very probable. The abrupt relief turns water the principal erosion agent, at work with an enormous sediment transportation capability. This fact is responsible for alluvial phenomenon that historically occur in Madeira Island, characterized by a flow of large amounts of sediments transported by water, concentrated in a small period of time (few hours), which may be extremely damaging. The alluvial episodes are well documented in historical reports [1, 2 and 3] and are triggered by a heavy rain episode that flood the river courses and allow the accumulated sedimentary material to be dragged and transported downstream with incredible force and velocity, dragging everything along the way. Alongside with the alluvial episodes is the formation of landslides [4, 5, 6 and 7], that usually nourish the alluvia with sediment material, while producing per se several material damages and sometimes life casualties. The most recent alluvium episode occurred on February 20 th , 2010, affecting mostly the hydrographic basins of Tabua, Ribeira Brava, São João, Sta. Luzia and João Gomes (Figure 1), all in the southern slopes of the island. Besides the landslides and the alluvium episode, the human figures are: 42 casualties, 250 injured and 600 homeless. This extreme event was triggered by a heavy rain episode, with rainfall values reaching twice the double of the monthly average: between 6 a.m. and 11 a.m. of the abovementioned day: 108 mm at Funchal station and 146 mm at the Pico do Areeiro station. A team with broad expertise was designated to conduct several studies in order to characterize this extreme event; some members dealt with the description of the precipitation, others with geological, geotechnical and hydraulic aspects, while others mapped the occurred landslides and alluvial. A crucial information for the aftermath evaluation was the identification of the landslides and the quantification of the sediment volume dislocated during the event. This information is vital for remediation plans and procedures that are being prepared to mitigate future similar events from being so much damaging. 2. METHODS The affected area comprises five hydrographic basins, totalizing an area of ca. 100 km 2 . This analysis would be impracticable to perform without automatic landslide recognition, which was conducted using remote sensing techniques.