An evaluation model for landslide and debris flow prediction using
multiple hydrometeorological variables
Abstract
Landslide and debris flows are typically triggered by rainfall-related
weather conditions, including short-duration storms and long-lasting
rainfall. The critical precipitation of landslide and debris flow
occurrence is different under various hydrometeorological conditions. In
this study, the daily hydrological states were evaluated by the SWAT
model, and the trigger sensitivities of different daily hydrological
variables were assessed with 50 days recorded landslide and debris flows
between 2010 and 2013. Based on modeled wetness states, the event days
were divided into LLR-trigger event days (long-lasting rainfall) and
SDS-trigger event days (short-duration storm) with six determinate
criteria. The landslide and debris flow prediction model was built using
nine hydrometeorological variables and the predictive performance was
tested with simulated data from 2010 to 2012. The results suggest that:
Historical hydrological variables and their development provide
important information for triggering debris flows, though rainfall is
the most important factor for triggering debris flows. The landslides
and debris flows in the selected subbasins region are triggered on 33
days by LLR and on 17 days by SDS. Specifically, LLR type landslide and
debris flow account for a large proportion in July, while SDS type
landslide and debris flow occur more frequently in September. The
prediction model with the AUC value of 0.85, can capture most of the
landslide debris flow. The temporal distribution of the two
triggering-event predicted by the model is consistent with the annual
distribution of precipitation. Besides, there are spatial variations of
the specific trigger types in the different subbasins, which attribute
to the different land cover. Despite some uncertainty, this study
thereby provides an idea of improving the landslide and debris flow
prediction model.