Basic HTML Version
Understanding the spatial and temporal occurrence of landslides using satellite and airborne technologies: Papua New Guinea
3
Fig. 2|
Regionally-averaged, monthly rainfall accumulations and monthly landslide frequency for (a) La Niña years (La Niña mean
monthly rainfall calculated based on 13 La Niña years recorded between 1970 and 2009), (b) El Niño years (El Niño mean monthly rainfall
calculated based on 11 El Niño years recorded between 1970 and 2009) and (c) ENSO neutral years (calculated based on 16 ENSO neutral
years recorded between 1970 and 2009). GPCC RMM = the GPCC regional mean monthly rainfall calculated on all years (1970 to 2009; 40
years), shown for comparison.
Although broad-scale relationships between rainfall and landslide occurrence were ascertained using the GPCC
rainfall climatology data, the detailed rainfall patterns associated with individual landslides could not be assessed
due to the temporal and spatial resolution of the datasets. This precluded the development of rainfall thresholds,
which formthe basis ofmany empirical landslidemodels used in early warning/forecasting systems. It was, therefore,
necessary to find an alternative source of rainfall data with higher spatial and temporal resolution to address this
issue. Satellite-based precipitation algorithms were considered for their extensive (spatial) and consistent (temporal)
data coverage. Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA; Huffman
et al., 2007) data, available at a spatial resolution of 0.25° x 0.25° and a daily temporal resolution (from January
1998 to the present), were used to examine
the rainfall characteristics which result in
landslides, over different temporal periods.
Daily rainfall time series were generated
for the 100 days preceding each landslide
that occurred between 1998 and 2009. The
different rainfall characteristics that can
result in landslides are illustrated in Fig. 3.
Given the range of uncertainties and
variability within both the rainfall and
landslide datasets, a method to develop
probabilistic rainfall thresholds for landslide
initiation was used. This involved a combined
approach based on the ‘multiple time
frames’ method (Frattini et al., 2009; Zêzere
et al., 2005; Fuhrmann et al., 2008) and
Bayesian statistics (Berti et al., 2012). Using
the ‘multiple time frames’ approach, rainfall
accumulations and mean rainfall intensities
were calculated for a number of rainfall
event durations (5, 10, 15, 30, 45, 60, 75 and
90 days). This was completed for both rainfall
events preceding historical landslides and
for rainfall events where no landslides were
recorded. Magnitude-frequency distributions
for rainfall events that resulted in landslides
were compared against the magnitude-
frequency distributions of non-landslide-
triggering rainfall events. The relative
Fig. 3|
Cumulative rainfall curves for a sample of landslide-triggering events
(LTE), illustrating the different rainfall characteristics which can lead to
landslides in different parts of PNG and at different times of the year. The month
of each landslide occurrence is shown on the top right of each plot. Horizontal
arrows represent different rainfall event durations which could have been
responsible for the landslides initiation.