Ground-Truth based Performance Assessment of Landslide Susceptibility Mapping: A Case Study after Elazig Earthquake (24 Jan 2020, Mw 6.8), Turkey

Abstract

Quality assessment for landslide susceptibility maps (LSMs) is essential to increase their usability. Ground-truth based quality assessment is crucial for performance evaluation of LSM production methods. Landslides triggered by earthquakes can be employed for this purpose as they occur frequently throughout the active seismic regions of the world. After an earthquake occurred in Elazig, Turkey on 24 Jan 2020 (Mw 6.8), several landslides were activated in the mountainous parts. Here, the performances of two state-of-the-art machine learning methods, i.e., random forest (RF) and multi-layer perceptron (MLP), were investigated using the activated landslides. The landslide inventory was derived by using pre- and post-event aerial photogrammetric datasets; whereas only the pre-event datasets were utilized for LSM production. The inventory includes activity classes of inactive (L1) and active mass movements (L2), areas having new active zones occurred after 2018 inside the existing landslide (L3), and new activity after 2018 (L4). The L1 and L2 type landslides were used in model training, while L3 and L4 type landslides triggered by the earthquake were used for validation. The area under curve (AUC) values and volumetric change maps obtained from the pre- and post-event datasets were also used for the performance assessment. Furthermore, only a part of the study area was used for model training while the LSMs were produced for the whole area. The results demonstrated that RF exhibited higher classification accuracy (AUC = 0.93) than MLP (AUC = 0.87); and accurate LSMs could be produced by using a sub-part of the basin for training.