Enhancing the reliability of landslide early warning systems by machine learning
- PDF / 1,980,092 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 72 Downloads / 178 Views
Hemalatha Thirugnanam I Maneesha Vinodini Ramesh I Venkat P. Rangan
Enhancing the reliability of landslide early warning systems by machine learning
Abstract This paper submits a report on the effective adoption of machine learning algorithms for enhancing the reliability of rainfall-induced landslides. The challenges involved in the design of reliable landslide early warning systems (LEWS) and the datadriven context for overcoming these challenges have been presented. The operation of LEWS is explained using the chain of five major components (i) Data collection, (ii) Data transmission, (iii) Modelling, analysis and forecasting, (iv) Warning, and (v) Response. Failure of any of these major components of the LEWS will break the chain of operation of LEWS and the ensued consequences of each component failure are reviewed. Inferences drawn from the analysis of the reliability measures incorporated in 12 LEWS deployments across a dozen locations around the world are also presented. Based on the investigations from 12 LEWS and the real-world experience, we identified that an alternate solution is required for ensuring the reliability of LEWS, especially during disaster scenarios when warnings are crucial, but data availability is a constraint. We recognized that machine learning algorithms can provide an alternate solution and in this paper, we have discussed two machine learning approaches nowcasting and forecasting for enhancing the reliability. Both the algorithms employ historic data of the landslide monitoring parameters to learn the changes materializing in slope leading to landslide incidences. The learned knowledge is used to nowcast and forecast the real-time and future conditions of the slope from the real-time landslide monitoring parameters. In terms of ensuring reliability, (i) Nowcasting algorithm provides an alternate solution if either the Data collection component or Data transmission component of a LEWS fails. (ii) Forecasting algorithm provides extra lead-time for early warning and solves the problem of less lead-time during early warning process. The breakthrough is even when the realtime landslide monitoring parameters are not available for various reasons, these algorithms take the minimal input of rainfall forecast information for nowcasting and forecasting thus restoring the broken chain of operation of LEWS. Keywords LEWS . Nowcasting . Forecasting Introduction and related works Landslides are one of the natural catastrophic hazards across many parts of the globe. The Asian continent ranks high in the spatial distribution of landslides, (Froude and Petley, 2018) with emerging global landslide hotspots (Petley 2012). From the recent dataset (Froude and Petley 2018) of non-seismic global landslides, over the period from January 2004 to December 2016, a total of 55,997 fatalities were recorded, in 4862 distinct landslide events. Many countries and research organizations have instituted LEWS, using various techniques, for the generation of early warning notifications of imminent landslides. Despit
Data Loading...
