USING PERSISTENT HOMOLOGY TO CREATE A LANDSLIDE INVENTORY FOR LUZERNE COUNTY, PENNSYLVANIA, USA

Landslide susceptibility is the measurement of the occurrence probability of landslides under certain geo-environmental conditions, and landslide susceptibility models/maps are considered the first steps toward determining landslide hazard and risk for an area. Susceptibility models are created by using the location of landslide events – compiled in landslide inventories – to assess their spatial distribution using physical-, knowledge-, or data-driven methods, with the latter being preferred as such methods can be utilized for expansive regions, and allow for the exploration between landslide frequency and geo-environmental and/or anthropogenic variables (i.e., slope, landcover, aspect, rainfall, etc.). Data-driven creation of a susceptibility map requires a landslide inventory to be created for the region. Landslide inventories for the study area of Luzerne County, Pennsylvania, such as the landslide inventory created by Karimi et al. 2019 for Northeastern Pennsylvania, typically use manual landslide detection methods to identify the size and location of landslides. Landslide inventories that are created by using manual detection methods require considerable amounts of time, money and personnel to complete. A semi-automated/automated detection model would significantly increase efficiency, reduce costs, and decrease personnel needed to compile the inventory. We propose that a topological tool - persistent homology – could be used to create an automated landslide detection method using points gathered from a ridgeline detection algorithm, and various dimensions. For the purposes of this study, ‘dimension’ refers to an individual data attribute (i.e., latitude, longitude, elevation, slope, etc.). Persistent homology would allow for the identification of persistent patterns that are associated with landslides, which in this case would be holes generated by a recognized pattern along the perimeter of a landslide. Previous work has shown that automated landslide detection can be an efficient method for landslide detection. This study presents a landslide inventory for Luzerne County, Pennsylvania by using persistent homology, an automated landslide detection method, to create a landslide inventory that is tested against the manually detected inventory generated by Karimi et al., 2019.