Landslide Susceptibility Mapping for South India with Dense Neural Network using Rainfall and Satellite Data-Driven Novel Dataset

Landslides are tragic events that come in short notice and take countless lives. This disaster is in desperate need of an early warning system. The proposed system aims to predict landslides beforehand in hopes of forming proper evacuation plans. A novel landslide dataset for South India was created by integrating weather and spatial data from Google Earth Engine and Geological Survey of India. The factors associated with spatial units include rainfall, NDVI and slope. A Dense Neural Network (DNN) was trained on this dataset in order to predict risk factor and thus plot the Landslide Susceptibility Map (LSM). Additionally, a Bi-LSTM model was trained on Wayanad”s rainfall data in order to predict the same. Thus, the primary objective is to predict rainfall using the Bi-LSTM model and use this as data fed into the DNN to plot LSM for upcoming dates. This would help us understand the landscape and probability of landslides of the future.