Statistical analysis of the landslides triggered by the 2021 SW Chelgard earthquake (ML = 6) using an automatic linear regression (LINEAR) and artificial neural network (ANN) model based on controlling parameters

This study uses automatic linear regression (LINEAR) and artificial neural network (ANN) models to statistically analyze the area of landslides triggered by the 2021 SW Chelgard earthquake (ML = 6) based on controlling parameters. We recorded and mapped the number of 632 landslides into four groups (based on the Hungr et al. 2014): rock avalanche-rock fall, debris avalanche-debris flow, rock slump, and slide earth flow-soil slump using remote sensing method, satellite images (before and after the earthquake), and field observation. The spatial distribution of landslides showed that the highest values of the landslide area percentage (LAP %) and of the landslide number density (LND, N/km2) occurred in the northern part of the fault on the hanging wall. The ANN models with R2 = 0.51–0.80 provided more accurate predictions of landslide area (LA, m2) than the LINEAR models, with R2 = 0.40–0.61 using multiple parameters. The LINEAR models revealed that the most influential controlling parameters for landslides were the topographic factors and ANN models showed that seismic parameters are effective on the coseismic landslides (e.g., the distance from the epicenter on the rock slumps; the PGA on debris avalanches- debris flow; the distance from the rupture surface of the fault and Ia on the rock avalanches-rockfall and slide earth flow-soil slump). Therefore, the classification of coseismic landslides can be helpful for predicting the LA more accurately and better understanding the failure mechanism.