Key points are not available for this paper at this time.
Landslides are considered to be calamitous natural hazards commonly recurring in the Indian Himalayas. Majority of landslides are induced by prolonged or heavy rainfall. Rainfall forecasting helps in identifying the precipitation conditions responsible for landslide occurrence. The proposed research work provides the performance comparison of various machine learning algorithms such as linear regression, back propagation neural network (BPNN), support vector regression (SVR) and long short term memory network (LSTM) used to forecast rainfall which can be compared with the rainfall thresholds to predict landslide occurrence. The analysis is performed using antecedent rainfall data obtained from Narendra Nagar, a small town in the Tehri Garhwal district of Uttarakhand for the period of 1901-2015. Owing to the limited predictability of instantaneous state of the weather, daily rainfall observations are aggregated into monthly indexes. The proposed algorithms use preprocessing techniques followed by data normalization to increase the accuracy of forecasting models. The developed models have the ability to predict rainfall intensity one month in advance or for a specific month of the upcoming year depending upon the dataset used. The study concludes that the BPNNs are able to outperform and provide optimal inferences stating the aptness of artificial neural networks (ANNs) in estimating rainfall and hence predicting the possibility of landslide occurrence well in advance. The study is conducted explicitly for regions highly vulnerable to landslides near Narendra Nagar but may be implemented to any landslide prone area.
Srivastava et al. (Mon,) studied this question.