Machine Learning For Nowcasting And Forecasting Landslides From Real-world IoT Data

With the advancements in sensor technology, communication and computing techniques, it is expected that disasters like landslides can be forecasted with higher accuracy than before. WSN and now recently IoT have facilitated scalability, fault tolerance, effective power consumption and ubiquitous computational abilities in sensor devices and facilitate data availability anywhere. These systems integrated with learning algorithms and knowledge discovery of the data from these devices can be of great use in life saving disaster early warning applications. Here, we discuss an efficient method for nowcasting and forecasting the real-time and 24-hours ahead conditions of the slope. The data from Amrita Vishwa Vidyapeetham's fully functional system using Wireless Sensor Networks (WSN) equipped with geophysical and geo-technical sensors is used for this purpose. The WSN system collects landslide triggering vital parameters such as rainfall, moisture, pore-water pressure, slope movements and earth vibrations. The key idea is to discover the interrelationships existing between the vital triggering parameters using mathematical models and machine learning algorithms. With the help of these models and the learned knowledge, we are able to (i) assess the stability conditions of the slope vulnerable to landslide with respect to rainfall and antecedent rainfall conditions, (ii) provide nowcasts even during times like disaster when data availability is a constraint, (iii) provide slope stability forecasts from rainfall forecast information. We also discuss how this system and the learning algorithms can be scaled to monitor other disasters like flood and drought.