Remote sensing and optimized neural networks for landslide risk assessment: Paving the way for mitigating Afghanistan landslide damage

Abstract

Landslides caused by mega earthquakes and other extreme climate change pose a major threat to lives and infrastructure. However, the lack of a detailed and timely landslide inventory and relevant risk assessment attributable to ongoing conflicts limits the effective prevention measures in Afghanistan. This study presents the first landslide inventory covering the whole nation of Afghanistan from 2015 to the present utilizing Google Earth Pro imagery and manual interpretation. Based on this inventory of 3,260 mapped landslides, we analyzed the distributional characteristics of landslides in Afghanistan and conducted a risk assessment that included landslide susceptibility and hazard, and vulnerability of the bearing areas. The existing regional studies attest to the accuracy and reliability of the inventory, and the results of the risk assessment using the optimized neural network method in this study are well validated. This study can provide a good database for the Afghan government to carry out relevant pre-disaster warnings and post-disaster reconstruction, which can help to delineate hotspots where landslides may occur, and reduce potential economic losses and human casualties from future landslides.

Keywords

• Landslide
• Afghanistan
• Risk assessment
• Remote sensing
• Neural networks