Data-Driven Landslide Nowcasting at the Global Scale

Landslides affect nearly every country in the world each year. To better understand this global hazard, Landslide Hazard Assessment for Situational Awareness (LHASA) model was developed previously. LHASA version 1 combines satellite precipitation estimates with a landslide susceptibility map to produce gridded of potentially hazardous areas from 60° North-South 3 h. categorizes world’s land surface into three ratings: high, moderate, and low hazard single decision tree that first determines if last seven days rainfall were intense, then evaluates susceptibility. 2 has been data-driven approach. The replaced collection explanatory variables, two new dynamically varying quantities added: snow soil moisture. Along antecedent rainfall, these variables modulated response current daily rainfall. In addition, Global Catalog (GLC) supplemented several inventories rainfall-triggered events. These factors incorporated machine-learning framework XGBoost, which trained predict presence or absence landslides over period 2015–2018, years 2019–2020 reserved evaluation. As result improvements, nowcast twice as likely occurrence historical 1, given same false positive rate. Furthermore, shift probabilistic outputs allows users directly manage trade-off between negatives positives, should make useful greater variety geographic settings applications. retrospective analysis, ran domain 5 years, results compared. Due importance faults 2, nowcasts would be issued more frequently some tropical countries, such Colombia Papua New Guinea; at time, placed less emphasis on arid regions far Pacific Rim. provides real-time view stakeholders.