Title: Modelling Rain-induced Landslides by Integrating Cellular Automata(CA) and Artificial Neural Network (ANN) Frameworks: The Case of Itogon, Benguet
Abstract
In the mountainous landscape of the Philippines, landslides continue to wreak havoc to life and property. The impacts of landslides, however, can be minimized by assessing the level of vulnerability within the community and preparing inhabitants for the occurrence of the event. This study presents an integration model for cellular automata (CA) and artificial neural network (ANN) frameworks to simulate and visualize rain-induced landslides in Itogon, Benguet. The novel model named LandSCANN (landslide simulation using CA-ANN) is capable of projecting landslides utilizing nonlinear properties learned from the identified landslide drivers. Several optimizations revealed a 9x18x1 low-cost network architecture with a learning rate of 0.01 and a momentum factor of 0.9 converging after 295 epochs. The generated landslide susceptibility index (LSI) had an accuracy of 87.62%. The modelled LSI, together with the flow direction layer, was then treated as a transition rule in the CA model. The calibrated automaton was used to create hypothetical landslides where landslide propagation was initialized in areas with an LSI of at least 0.99. Layout maps were crafted to effectively visualize each hypothetical landslide with information on potentially affected built-up areas. Consequently, the visualization may help the locals of Itogon in identifying and minimizing landslide vulnerability.