Towards an automatic landslide mapping tool based on satellite imagery and geomorphological parameters. A study of the Itogon area (Philippines) after Typhoon Mangkhut

Landslides and floods driven by typhoon and monsoon rainfall cause thousands of fatalities and millions of pesos in damage to infrastructure and commerce in the Philippines each year. The Philippines accounts for 46% of rainfall-triggered landslides in SE Asia, although it represents only 6% of the land area (Petley, 2012).

Despite their relevance, landslide inventories are very scarce in the Philippines, and most of them are point-based inventories, so lacking landslide magnitude. This makes it difficult both to assess their magnitude-frequency relationships (major component of hazard assessment) and to provide landslide sediment delivery rates to the river network (needed for better prediction of channel morphodynamics, flood risk and reservoir management), which is one of the main goals of the SCaRP project (Simulating Cascading Rainfall-triggered landslide hazards in the Philippines), funded under Newton Programme (UK Research and Innovation).

Manually mapping landslides to obtain polygon-based landslide inventories in areas affected by RILs (Rainfall Induced regional Landslide events) is a time-consuming task, which is often not affordable for the authorities in terms of resources and time. Meanwhile, automatic methods to map landslides based on satellite imagery have broadly improved during the last decade (e.g.: Alvioli et al 2018).

The city of Itogon (Benguet, Luzon) and its surroundings was hit by typhoon Mangkhut in September 2018, which triggered thousands of landslides, including a fatal one that killed over 70 miners. We selected a test area of 135 km², with a high density of landslides.

The objective of this work was twofold: 1) to characterize the geomorphological features of the landslides that occurred in the area of Itogon due to the passage of Typhoon Mangkhut, 2) to analyze the potential of automatic tools to map landslides from satellite imagery.

A total number of 1100 shallow landslides and flows were manually mapped, with areas ranging from tens to tens of thousands of m².  An automatic pixel-based approach (developed within H2020 HEIMDALL project and called Slidex)  was tested, which relies on a Random Forest classification using Sentinel-2 bands and a set of radiometric indices. The algorithm was trained over several regions (e.g. Japan, Sierra Leone) and applied to the Philippines. The results suggest that the change in land cover is the best indicator to identify landslides automatically, though the efficiency of the tool was improved by including geomorphological parameters such as slope and minimum area affected.

 

Alvioli, M., Mondini, A. C., Fiorucci, F., Cardinali, M., & Marchesini, I. (2018). Topography-driven satellite imagery analysis for landslide mapping. Geomatics, Natural Hazards and Risk, 9(1), 544–567. https://doi.org/10.1080/19475705.2018.1458050

Petley, D. (2012) Global patterns of loss of life from landslides. Geology, 40(10), 927-930

How to cite: Abancó, C., Bennett, G., Briant, J., and Battiston, S.: Towards an automatic landslide mapping tool based on satellite imagery and geomorphological parameters. A study of the Itogon area (Philippines) after Typhoon Mangkhut, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-17940, https://doi.org/10.5194/egusphere-egu2020-17940, 2020