Assessing Landslide Susceptibility in Active Tectonic Zones using Convolutional Neural Networks (CNN)

Assessing landslide susceptibility is vital in mitigating natural disasters, particularly in zones with active tectonics. This paper will discuss a novel approach based on a CNN architecture in a U -Net designed to assess landslide susceptibility in these zones of active tectonism. The model thus takes advantage of spatial data sources from remote sensing and GIS and multi-coupled geospatial factors such as slope, elevation, rainfall, and tectonic activity in a multivariate framework for predicting landslide-prone areas. The U-Net CNN architecture is used as it captures the local and global spatial patterns and results in very precise susceptibility maps. The selected historical data about landslides was then fed into the trained and validated model to predict potential landslide occurrence zones, indicating high accuracy in susceptibility mapping. The CNN-based U-Net approach suggests a higher order of precision and spatial resolution than the conventional machine learning approach. This work portrays the capabilities of deep learning approaches in developing landslide risk assessment with much-improved precision. It thus offers an essential tool for disaster risk management along with urban planning for tectonically active regions.