Small size but densely distributed: Insights from a LiDAR-based manual inventory of the recent earthquake-induced landslides case in Japan

Individually delineated landslide inventories are essential in analyzing post-earthquake-induced landslides (EIL) hazard assessments, particularly examining statistical correlations between landslides (e.g., frequency and size) and physical parameters. Despite rapid advances in remote sensing technology, previous recorded EIL inventories still have limitations in carrying out fine quality inventories, mainly due to limitations in delineating individual landslides manually over large areas by low-resolution satellite images. To be specific, fine quality inventory requires the ability to detect landslide scars and deposits separately over whole affected areas, recognizing smaller landslide sizes (<10³ m²) under canopies, as well as avoiding amalgamations, i.e., a combination of several individual landslides in a single polygon, which can lead to severe distortion of landslide statistics. The latest technology from LiDAR-Digital Terrain Model (DTM) allows geomorphologists to manually delineate landslides precisely, but most studies had only focused on deep-seated landslides. Thus, the main objective of this study was to delineate the recent EIL based on LiDAR-DTM visualization over whole landslide-affected areas and test preliminary statistics between our manual LiDAR-based inventory (MLI) with automatic aerial-based inventory (AAI) in the same areas, in addition to NASA’s global EIL database.

We manually delineated the recent landslides affected by the 2018 Eastern Iburi earthquake in the Atsuma basin in Hokkaido within an area of 266 km², accounting for about 90% of the total area affected by landslides. Shaded relief derived from LiDAR-DTM (0.5 m), and aerial photograph (0.2 m) were used to identify landslide morphometrics. AAI collected in the same study area (Kita, 2018) was used to compare with MLI. As a result, our MLI was able to detect a total of 17,160 landslides (total landslide area: 27.5 km²) while the automatic AAI was only 4241 landslides (total landslide area: 33 km²), probably because our MLI was able to recognize more small landslides and separate individual landslides from amalgams. The mean landslide density for MLI is four times greater (64 landslides/km²) compared to AAI (16 landslides/km²), also considered the densest landslide inventory recorded so far in 20 years based on NASA's global EIL inventory database. Based on the binned frequency area distribution (FAD), MLI has a power-law exponent (β) of 3.4 and a rollover point of 800 m², whereas AAI is 2.7 and 3×10³ m², respectively, probably because AAI's inventory overestimates its delineation by inserting channels and depositional regions in the delineated polygons. Compared with all global EIL inventories (mean β: 2.4), the value of the MLI was found to be larger, indicating that the Iburi EIL is the smallest size EIL so far in history (50% landslides are smaller than 10³ m²), but very dense. Our findings suggest that MLI might reveal hidden unexpected statistics of the number and size of EILs, including exposing smaller landslides under the canopy and splitting amalgams.