Enhanced Mapping of Rainfall Induced Landslide Susceptibility Using a Deep Feedforward Neural Network with Soft Computing

Abstract

The presented study attempted to propose enhanced rainfall-induced landslide susceptibility mapping method by using the Deep Feedforward Neural Network (DFNN) which is developed for analysis the non-liner feature detection in landslide susceptibility analysis. To evaluate our approach, a comprehensive dataset of triggering factors was compiled, encompassing historical landslide occurrences with total of 107 records, rainfall data, geological information, seismicity, human-activities, and topographic attributes. Through rigorous training and testing procedures, the DFNN demonstratedsuperior ability for generalization and superior performance. The effectiveness of the selected method is demonstrated on the data from the Zanjan County, known for its diverse geographical, geological, and hydrological characteristics, which are pivotal factors in mapping of landslide susceptibility. Results showcased a substantial enhancement in the accuracy of mapping of rainfall-induced landslide susceptibility for the Zanjan County, which is compared with benchmark learning classifiers. According to the results of the study, it appeared that the northeastern and southwestern area of the Zanjan County can be deemed to have a high to very-high risk of landslide occurrence, which is validated via benchmark classifiers. The western part of the Zanjan County was observed to have a very low to low risk.