2185
A site specific early warning system for rainfall induced landslides
Utilisation d’un site spécifique pour l’élaboration d’un système d'alerte rapide pour les instabilités de
pente induites par les pluies.
Harris S., Orense R.
Department of Civil and Environmental Engineering, The University of Auckland, Auckland, New Zealand
Itoh K.
National Institute of Occupational Safety and Health, Tokyo, Japan
ABSTRACT: An early warning system (EWS) to warn users of imminent landsliding caused by rainfall has been developed. The
EWS is deterministic, based on a pre-determined failure mechanism present at a specific site. A prototype of the EWS was developed
for a roadway embankment located in Silverdale, New Zealand. Prolonged rainfall caused a landslide at the site in 2008. Soil debris
from this landslide event almost obstructed a major highway, which could have been potentially dangerous to motorists as well as
causing major delays to the Auckland roading network. Volumetric water content sensors were installed at various depths and
locations along the same cross section of the site. A 2D finite element model was used to replicate the response of the sensors to
rainfall, using monitored rainfall events as an influx in the model. Next, a limit equilibrium analysis was used to obtain the factor of
safety against slope failure for each time step in the finite element model. An artificial neural network was then trained to predict this
factor of safety using the sensor readings as inputs. Thus, the factor of safety of the slope can be predicted in real time. This predicted
factor of safety forms the basis of the EWS.
RÉSUMÉ : Un système d'alerte précoce (SAP) pour avertir les utilisateurs de glissements de terrain provoqués par des pluies
imminentes a été développé. Le SAP est déterministe, basée sur les mécanismes de rupture pré-déterminés présents sur un site
spécifique. Un prototype du SAP a été développé pour un remblai de la chaussée située à Silverdale en Nouvelle-Zélande. Des pluies
prolongées ont causé un glissement de terrain sur le site en 2008. Les coulés de sol engendrées par ce glissement de terrain ont
presque obstrué une route importante, ce qui aurait pu être potentiellement dangereux pour les automobilistes ainsi qu’être à l'origine
de retards importants sur le réseau routiers d’Aucklande. Des capteurs volumétriques de teneur en eau ont été installés à des
profondeurs différentes et à des emplacements variés le long de la section transversale du site. Un modèle par éléments finis 2D a été
utilisé pour reproduire la réponse des capteurs aux précipitations, en utilisant les données expérimentales de comme données d’en
trée. Ensuite, une analyse d'équilibre limite a été utilisée pour obtenir le facteur de sécurité pour la stabilité de la pente pour chaque
pas de temps. Un réseau neuronal artificiel a ensuite été formé pour prédire ce facteur de sécurité en utilisant les relevés du capteur
comme modèle. Ainsi, le facteur de sécurité de la pente peut être prédite en temps réel. Ce facteur de sécurité prévu est à la base du
SAP.
KEYWORDS: rainfall, landslide, artificial neural network, early warning system
1
INTRODUCTION
As a means to mitigate the risk of rainfall induced landslides
which cause millions of dollars’ worth of damage each year in
New Zealand (NIWA & GNS Science, 2010), an early warning
system (EWS) has been developed. A prototype of this EWS
was installed at a site in Silverdale, Northland, New Zealand.
Much of the damage which incurs from rainfall induced
landslides occurs in this region of New Zealand (NIWA & GNS
Science, 2009).
EWSs for rainfall induced landslides started as empirical
relationships which related the number of landslides in a given
region to the intensity and duration of rainfall events. Examples
can be seen in Dhakal & Sidle (2004), Keefer et al (1987) and
Caine (1980). As technologies have developed, focus on EWSs
has become more site specific. Current EWSs rely on measuring
parameters such as pore pressure and displacement at a given
site. Such EWSs are based on issuing an alarm when a
predetermined level of these parameters has been reached (Chae
& Kim, 2012; Intrieri et al., 2012). The EWS developed in this
research was required to return to the user a number related to
the possibility of failure, and also a timeframe for failure to
occur. To achieve this, volumetric water content (VWC) sensors
were installed at a variety of depths at the toe, mid-point and top
of the slope. A tipping bucket rain gauge was used to monitor
the intensity and duration of rainfall events. The fluctuations in
VWC recorded by the sensors were replicated in a finite
element model (FEM), using the recorded rainfall events as an
influx into the slope. Next, a limit equilibrium analysis was
used to determine the factor of safety (FOS) at each time step in
the FEM. Thus, a database was created which contained values
of the VWC as measured by the sensors at the site, and the
corresponding FOS. This database was then used to train an
artificial neural network (ANN). The ANN can thus predict the
FOS of the slope in real time, using sensor readings as an input.
The ANN can also predict the future FOS of the slope, using
rainfall forecasts for the site as an input. The trend of the
predicted FOS using the sensor data, and the future FOS
obtained according to the rainfall forecast, form the basis of the
EWS. Based on this information the user of the EWS can take
the required action; in this case, lowering speed limits and
putting detours in place.
1.1
Site and Soil Description
The site consists of a roadway embankment created from a cut
operation during the construction of State Highway One, which
runs parallel to the toe of the embankment. State Highway One
is a major arterial which services Auckland city. The slope
angle of the embankment is approximately 15°. A concrete dish
drain is located on a bench at mid-height of the slope. The site is
grassed, with some low height trees present. Debris from a
landslide which occurred at the site in 2008 following
prolonged rainfall almost crossed into the traffic lanes of State
