Assessing the Near-Future Behavior of a Landslide: Development and Preliminary Results of a Machine Learning Algorithm

The importance of monitoring hydrogeological instability has grown over the years, since landscape anthropization results in a rising number of people and structures in areas at risk of natural hazards. The recent development of highly sophisticated sensors and infrastructures has dramatically increased the amount and variety of collected data. This has the potential to enable more reliable and re-sponsive Early Warning Systems, but brings forth the need for the automation of data processing. The approach here proposed relies on the application of Machine Learning algorithms to landslide monitoring, leveraging the unprecedented amount of information through neural networks and innovative techniques from the field of Artificial Intelligence. The key concept involves the application of Machine Learning to obtain a prediction of the landslide behavior in the near fu-ture, which will enable the comparison with predefined thresholds. The prediction is generated by a neural network that learns how to correlate one state of the mon-itored site with the correct following state using the large number of examples at its disposal. The proposed method provides a time frame for the detection of dan-gerous events in the observed area, continuous real-time monitoring, and auto-mated alarm signaling. The procedure is currently being tested on a landslide lo-cated in Northern Italy, monitored since December 2018 with a system including 4 automatic modular underground monitoring system (MUMS) inclinometers, 2 barometers, and a total of 6 piezometers. The multi-parameter approach allowed to collect a considerable amount of information regarding the monitored site, im-proving the algorithm’s reliability and robustness.