PUBLICATIONS
Journal of Geophysical Research: Atmospheres
RESEARCH ARTICLE
10.1002/2013JD020881
Key Points:
• A flash flood/extreme precipitation
related natural persistent hazards
• New methods applied in climate
sensitivity of extreme precipitation
• Methods also applicable to 2010 type
U.S. flash floods
The devastating Zhouqu storm-triggered debris flow
of August 2010: Likely causes and possible
trends in a future warming climate
Diandong Ren1
1
Department of Imaging and Applied Physics, Curtin University of Technology, Perth Western, Western Australia, Australia
Abstract On 8 August 2010 in the northwestern Chinese province of Gansu, a rainstorm-triggered debris
Correspondence to:
D. Ren,
[email protected]
Citation:
Ren, D. (2014), The devastating Zhouqu
storm-triggered debris flow of August
2010: Likely causes and possible trends
in a future warming climate, J. Geophys.
Res. Atmos., 119, 3643–3662,
doi:10.1002/2013JD020881.
Received 13 SEP 2013
Accepted 13 FEB 2014
Accepted article online 18 FEB 2014
Published online 2 APR 2014
flow devastated the small county of Zhouqu. A modeling study, using a new multiple-phase scalable and
extensible geofluid model, suggests that the cause is an intersection of several events. These were a heavy
rainstorm, not necessarily the result of global warming, which triggered the landslide and followed a drought
that created surface cracks and crevasses; the geology of the region, notably the loess covering heavily
weathered surface rock; and the bedrock damage, that deepened the surface crevasses inflicted by the
7.9 magnitude Wenchuan earthquake of 12 May 2008. Deforestation and topsoil erosion were critical
contributors to the massive size of the debris flow. The modeling results underscore the urgency for a high-priority
program of revegetation of Zhouqu County, without which the region will remain exposed to future
disastrous, “progressive bulking” type landslides. Debris flows are more predictable types of landslides;
consequently, a series of “pseudo climate change” model experiments of future extreme precipitation events
are carried out using the Weather Research and Forecasting model, forced by temperature perturbations
from an ensemble of climate models. In a possibly future warmer climate, extreme precipitation events are
anticipated to be more severe, and this study has identified an atmospheric blocking pattern that might
produce future extreme precipitation events in the peri-Tibetan Plateau (TP) area (located to the northeast of
the TP). Importantly, observations from gravity field measuring satellites indicate that the larger geological
environment of this region also is becoming increasingly unstable.
1. Introduction
A debris flow on 8 August 2010 in the northwestern Chinese province of Gansu, was responsible for more
than 1000 deaths when it devastated the small county of Zhouqu. The future occurrence frequency of these
landslides therefore is of immediate practical concern. The geological and hydrological settings of the
Zhouqu landslides (Figure 1a) are vital to gaining a greater understanding of these events. The extreme
precipitation from the rain storm that triggered the landslides was caused by a midlatitude frontal system,
manifested as a shear zone at the 700 hPa level, rather than a closed surface low associated with the cold
front passage, due to the high average elevation (in excess of 3000 m) of the region. In the first part of this
article, the key question “Why was the August 2010 Zhouqu landslide so powerful?” is addressed by using the
high-resolution precipitation output from a numerical weather prediction model (the Weather Research
and Forecasting (WRF) model) [Skamarock et al., 2008] to drive an advanced landslide modeling system
(SEGMENT-Landslide) [Ren et al., 2008, 2009, 2011]. The model allows the spatial distribution of sliding
material to be estimated. Debris flow landslides are more predictable because of their strong correlation with
extreme precipitation [Ren et al., 2011]. Future extreme precipitation events can be examined using advanced
numerical weather prediction model, combined with a recently developed technique [Lackmann, 2013] for
perturbing the thermal background, to mimic a warming climate environment. This “pseudo climate change”
approach is adopted to estimate possible changes in precipitation in a 2060–2080 time frame. The
characteristics of future “Zhouqu” types of debris flows then can be determined from the landslide model.
Landslides are partly upscale processes, with localized disturbances enhancing instability by adding fluids or
reducing root reinforcement of preexisting weathered regolith and granular soil particles spreading on slopes in
regions predisposed to geohazards. Remote sensing observations over the past decade provide an opportunity to
examine the bedrock cracking conditions over the peripheral Himalayan region, which includes both Zhouqu
County and the areas affected by the 2008 Wenchuan earthquake. By identifying regions suffering steady mass
REN
©2014. American Geophysical Union. All Rights Reserved.
3643
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
(a)
(b)
Figure 1. (a) Image of the Zhouqu debris flows of 8 August 2010 after cessation. Source: NASA. (b) Inset shows the selected
WRF domains and the nested domain setup. The nesting ratio is 5, 4, 3, and 3, respectively. The innermost (fifth layer) is
90 m resolution. Domain 1 is about 16 km resolution. Color shades in Figure 1b are the shaded relief image of the 90 m
digital elevation map in Domain 5 (Y. Ping, personal communication, 2013). Zhouqu City is marked with a “star.”
gain/loss over the past decade and knowing the precipitation trends over the same period, it is possible to deduce
the sources of increasing/decreasing groundwater. Regions with increased partitioning of precipitation into
groundwater may be linked with a corresponding increase in bedrock crevasses. As the climate warms, extreme
precipitation events are likely to become more common [Trenberth, 1999], and future recurrence frequencies of
such events therefore must be estimated and included in the forcing of the landslide model.
2. Model and Methods
2.1. Model
Specifically, the landslide model used in this study is a version of the SEGMENT modeling system, referred to
as SEGMENT-Landslide [Ren et al., 2008, 2009, 2011]. SEGMENT-Landslide is used to investigate the Zhouqu
REN
©2014. American Geophysical Union. All Rights Reserved.
3644
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
debris flow of 8 August 2010, particularly
its cause and possible future preventative
actions. The Zhouqu landslides were
preceded by an extreme precipitation
event that occurred around midnight on
7 August 2010 (Figure 2). Both the
precipitation intensity, which peaked at a
rate of 77.3 mm/h near 104.42°E, 33.78°N,
and the total rainfall of 96.3 mm in 24 h
were the highest recorded in the
region since the May 2008 Wenchuan
(magnitude 7.9) earthquake. From a
longer perspective, the Zhouqu
rainfall event had a 20 year probability
of occurrence under the present
climatology as determined by, for
example, using a generalized Pareto
Figure 2. The daily precipitation time series for Zhouqu (33.875°N, distribution analysis [Coles, 2001; Balkema
104.375°E), for the period 1 January 1998 to 20 August 2010 (not every
and de Haan, 1974], considering ongoing,
day has precipitation). These are estimated from TRMM (3B42V6, microsignificant climate change. The hills
wave-IR mixed products) three-hourly precipitation. The last 2 months
around Zhouqu have a long landslide
are from rain gauge measurements. The left inset is a zoomed-in period
history [http://news.sciencenet.cn/
after the 2008 Wenchuan earthquake. The right inset shows the rainfall
histogram based on the landslide triggering rain event analyses prohtmlnews/2010/8/235921-1.shtm; Ma
posed by Ren et al. [2011]. Rain events with rainfall totals >30 mm can
and Qi, 1997; Bolt et al., 1975, p. 187].
trigger significant landslides in the Zhouqu region (see inset histogram).
However, this event is unique in its
Moreover, in that period, there are seven events with rainfall totals
unprecedented magnitude, involving
>60 mm, which therefore are rainstorms as intense as that which pre6 3
ceded the 7 August 2010 Zhouqu mudslides (the event indicated by the ~2.05 × 10 m of sliding material.
red arrow in the left inset). Thus, extreme precipitation alone does not
Because the landslide produced
explain the magnitude of the Zhouqu 2010 debris flow.
significant loss of life and great economic
cost, it has generated intense discussion
about the possible cause(s) of the slide. These include previous drought conditions that caused surface
crevasses, unprecedented intense precipitation, slopes loosened by the Wenchuan earthquake, an historical
earthquake (over a century before) and its leftover debris, and environmental consequences as population
increases resulted in the habitation of previously unoccupied locations. Climate warming often is seen by some
as the major cause, by contributing to the severity of the rainstorm [Intergovernmental Panel on Climate Change
Fifth Assessment Report (IPCC AR5), 2013, Chapter 14], whereas others argue that it was the recent drought,
which produced cracks in the soil mantle. In addition to its geological uniqueness, because it occurred so soon
after the 2008 Wenchuan earthquake, the presence of earthquake broken bedrock also is cited as a factor
contributing to the size of the landslide. A number of factors other than extreme precipitation therefore are
offered as responsible for magnifying the Zhouqu landslide to its unexpected great size [Ma and Qi, 1997;
Yu et al., 2010].
2.2. Method
To quantitatively investigate the relative importance of the various possible causal factors, the SEGMENTLandslide model was applied to the event. SEGMENT-Landslide is a fully three-dimensional, dynamical
landslide model that incorporates not only soil/rock mechanical properties but also the hydrological and
mechanical effects of vegetation on storm-triggered landslides. The SEGMENT-Landslide model is outlined in
the Appendix. The model requires a wide range of input variables such as land cover, land use, and geological
data, all of which were provided by Beijing Normal University. The digital elevation data were obtained from
the Shuttle Radar Topography Mission (http://srtm.mgs.gov/) at 90 m resolution (Figure 1b). To reproduce
historical landslides, precipitation forcing was derived from the Advanced Research WRF [Skamarock et al.,
2008] in a one-way nested mode with the innermost domain having the same horizontal resolution as the
digital elevation map (DEM), which is 90 m (Figure 1b). The multiple-level nesting approach circumvents the
necessity of dynamic [Pielke et al., 2012] or empirical downscaling procedures, in order to match the variant
REN
©2014. American Geophysical Union. All Rights Reserved.
3645
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
resolutions of model precipitation and topography. The selected SEGMENT-Landslide domain is only a
subdomain of the innermost WRF domain (33.66–34.06°N, 104.26–104.66°E). The output frequency of
WRF is set to be hourly. Surface biomass loading was provided by the Moderate Resolution Imaging
Spectroradiometer (MODIS) products [Zhao and Running, 2010; Zhang and Kondragunta, 2006]. A team
survey of the area also enabled a 300 m resolution vegetation mask to be produced. To investigate
possible mechanisms, several sensitivity experiments were made with a range of specified vegetation
conditions. Geologically, the hilly terrain of the region is composed mainly of metamorphosed
limestones, interspersed with altered clay layers. The ground surface rocks range from highly to
completely weathered. The weathered rocks date from the Paleozoic (primarily the Permian period)
and Mesozoic eras, the yellowish interbedded sandstone and siltstone date from the Silurian period,
and the gray limestones date from the Triassic period. The infiltration of rainfall through the
macropores, which are well developed in the soil and rock mass of the Zhouqu region, is critical for
slope stability. The hills intersect with canyons, where increased erosion occurs during the highly
regular rainy season.
For simulating future (2060–2080) landslide events, the WRF model is used in the same pseudo global
warming approach proposed by Lackmann [2013]. There is no simple means of anticipating how future
synoptic pattern changes might influence the frequency and severity of those extreme precipitation events
that trigger landslides. The experimental design for simulating future extreme precipitation events addresses
this issue by replicating the synoptic pattern that caused the 7–8 August precipitation, but imposing possible
future large-scale thermodynamic changes. While it is acknowledged that an identical synoptic pattern
will not occur in the future, it is reasonable to assume that a similar pattern could occur. Because the
model simulation is allowed to evolve dynamically for the duration of the synoptic event, the resulting
changes, brought about by imposing thermodynamic perturbations, can also induce dynamical changes.
Quantification of projected thermodynamic changes due to increased anthropogenic greenhouse gases is
accomplished using an ensemble of Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment
Report general circulation model (GCM) simulations of monthly temperature, sea surface temperature, and
mixing ratio values. Experiments using the A1B scenario are used. The August monthly fields were averaged over
a subset of five GCMs for which reliable temperature data are available at all vertical levels: the Bjerknes Centre for
Climate Research Bergen Climate Model version 2, the Centre National de Researches Meteorologiques Coupled
Global Climate Model version 3, the Institute of Numerical Mathematics Coupled Model version 3, the MaxPlanck
Institute European Centre/Hamburg version 5, and the third climate configuration of the UK Met Office Unified
Model (HadCM3). A two-decade average temperature change was then computed from the August monthly
averages from the five-member GCM ensemble for the periods 2000–2020 and 2060–2080. Similarly, sea surface
temperature, air temperature at 2 m, and all available variables on isobaric levels were averaged. These spatially
varying average fields were interpolated to the same 1° U.S. Global Forecast System (GFS) grid and used for the
initial and lateral boundary conditions in the control (2010) simulation.
The differences between the 2060s and 2000s decades were computed for each grid cell, and these changes
were subsequently added to the original GFS analyses. Given that the landslide event in question took place
in August 2010 and that a change is added that was calculated from a 60 year difference, the future
simulation can be interpolated as approximating the thermodynamic conditions around the year 2060 that
would arise from anthropogenic greenhouse gas forcing alone, under the A1B scenario. The same synoptic
weather pattern from August 2010 also characterizes the future simulation, and this fictitious date does not
correspond to a specific date in future GCM projections. To resolve the mesoscale evolution of this event, the
model domain features a larger outer domain with 16 km grid spacing, along with a 90 m nested innermost
domain within which SEGMENT-Landslide is actually run. Vertically, the model uses the default 28-level
configuration, with the model topset at 50 hPa. The Betts-Miller-Janjic convective parameterization scheme
was used only for the outermost domain only. All domains employed the WRF single-moment six-class
microphysics scheme, the Yonsei University planetary boundary layer scheme, and the National Oceanic and
Atmospheric Administration land surface model. Longwave radiation was handled with the rapid radiative
transfer model scheme, and shortwave radiation was handled with the Dudhia scheme. Initial and lateral
boundary condition data are from the GFS final analyses on a 1° latitude-longitude grid; simulations were
initialized at 00 UTC 4 August 2010 and run for 6 days. For the nested domains, the parent domain provides
lateral boundary conditions, and one-way nesting is used.
REN
©2014. American Geophysical Union. All Rights Reserved.
3646
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
3.Results
3.1. Reproducing the Present August
2010 Zhouqu Landslide
For reproducing the 8 August 2010
debris flows that occurred in Zhouqu
County (34.19°N, 104.41°E), spatially
distributed, high-resolution
precipitation rates are a prerequisite.
Because of the complex terrain around
Zhouqu (it lies in a northwest-southeast
oriented narrow valley. Surface
elevations to the west are generally
above 3000 m, i.e., ~800 hPa; Figure 1b),
precipitation is very patchy. In situ
measurements typically are unable to
identify the “hot spots” (i.e., locations
with extreme precipitation and may
have debris flow hazards). Remote
sensing precipitation products, such as
the satellite-based National Aeronautic
and Space Administration (NASA)
Tropical Rainfall Measuring Mission
(TRMM), are only of 0.25° × 0.25°
resolution, which is much greater than
the 90 m DEM. Nested runs of WRF, with
the innermost domain set at 90 m
resolution, are the best available choice
for the purpose. The model is run
for 6 days continuously, starting from
00 UTC 4 August 2010.
At around 06 UTC 7 August at 700 hPa
(and lower levels), there is a clear shear
zone of significant cyclonic vorticity.
Figure 3 shows the synoptic conditions
at 12 Z 7 August 2010, approximately
10 h before the storm peak. Figure 3a
Figure 3. Synoptic background of the Zhouqu extreme precipitation.
shows the National Centers for
NCEP/NCAR reanalysis of the (a) low-level (850 hPa) and (b) midlevel
Environmental Prediction/National
(500 hPa) streamlines, wind speeds (dashed lines; m/s), and relative vor5 1
Center for Atmospheric Research (NCEP/
ticities (color shades; 10 s ) at 12 Z 7 August 2010.
NCAR) reanalysis at 850 mb at 12 Z. At
500 hPa, the synoptic scale highs and lows form a saddle structure. The flow field brings cold and dry air to
the region of interest. At lower levels, the shear line is a convergence line (centered at 34°N, 104°E), and a
warm, moist air mass from the south is forced to ascend. This thermal stratification has a large amount of
convective available potential energy (~2540 J/kg at 12 Z 17 August 2010). From the circulation fields, it is
clear that there is a series of terrain forced meso-γ scale cyclones developing around Zhouqu. These features
are readily identifiable from the MODIS cloud top imageries at around 0635 UTC 7 August aboard the NASA
EOS Aqua satellite. River evaporation apparently contributed because the clouds as a group are aligned with
the valley direction. As these meso-γ scale cyclones developed, strong, local precipitation was produced. The
WRF simulated precipitation total amount in Region 2 during the 4 days leading up to the extreme
precipitation period was examined (figures not shown). The WRF captured the orographically induced
features in the spatial precipitation distribution. The prevailing winds were westerly, so the leeward side has
rain shadow areas. The comparison between WRF simulated and automatic station measured precipitation
rates (mm/h) also are satisfactory. Among the 10 stations within this small area, for those that recorded
REN
©2014. American Geophysical Union. All Rights Reserved.
3647
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Figure 4. The WRF simulated precipitation total amount in Domain 3 during 00 Z 7–8 August 2010. The WRF captured
the apparent orographic effects in the spatial precipitation distribution. Prevailing winds are westerly, so the leeward
side has rain shadows. The inset is a comparison between WRF simulations and automatic station measured precipitation rates (mm/h).
significant precipitation, all agree well with observations (not plotted for clarity). At stations with weaker
precipitation, phase (timing) and magnitude errors become significant.
For the storm that directly triggered the Zhouqu landslides, the WRF simulated precipitation spatial patterns
in Domains 1 and 2 agree well with the TRMM observations, but are generally more concentrated as the
magnitude of TRMM precipitation is less than the WRF simulations. Figure 4 is the WRF simulated total
precipitation amount during the period 00 Z 7 August–00 Z 8 August 2010, in Domain 3. The patchiness of
precipitation results from the atmospheric conditions and the complex topography of the region. The inset of
Figure 4 shows observed precipitation rates at a station located at 104.42°E, 33.77°N. Although the phase
error in the timing of precipitation is satisfactory, WRF has a >10% underestimation of the peak precipitation
rate. It is assumed that this underestimation is systematic and does not affect our conclusions concerning the
sensitivity of storm-triggered landslides to climate change.
From Figure 4, the precipitation around Zhouqu is not the strongest over the region. There are however
significant antecedent precipitation events on 16, 21, and 24 July and 4 August, respectively. Consequently,
the soil moisture is high. With the high spatial and temporal resolution precipitation rates provided by WRF,
Figure 5a shows the SEGMENT-Landslide simulated unstable areas, as indicated by the maximum obtainable
surface sliding speed. Under the current vegetation regime, the most significant scar is that near the
Sanyanyu Valley (33.81–33.87°N, 104.36–104.42°E). This particular sliding (Figure 6) is a characteristic
“progressive bulking” type of debris flow [Iverson, 1997]. The accumulation area spreads up to 3500 m
elevation, in a fan shape with the fan “handle” extending down to the Bailongjiang River. The surface runoff
essentially is clear water above the 3500 m elevation contour, but at lower elevations, it gradually becomes
turbid and entrains small stones and coarse granular material into the slide streams. These creeks are
usually dry except during rainy periods. Figure 7 is an enlargement of the Sanyanyu gully, showing the
surface elevation changes at 2 times in the sliding process: the beginning (Figure 7a) and the cessation
(Figure 7b). At cessation, the areas indicated by two red arrows have little elevation change, despite the
massive total mass in the slide. They acted like a pathway for the sliding materials at higher elevations.
REN
©2014. American Geophysical Union. All Rights Reserved.
3648
Journal of Geophysical Research: Atmospheres
(a)
10.1002/2013JD020881
(b)
Figure 5. A detailed comparison of the unstable areas identified by the landslide model. These are areas for which model
landslide speeds (m/s) exceeded the threshold value. (a) Current vegetation. (b) All vegetation removed. Under current
vegetation conditions, only the Sanyanyu area is unstable. When the vegetation is removed, there are many additional
unstable areas. Moreover, the landslide flow magnitudes are larger than for the vegetated case.
For example, at point A, there is a break in the slope where some of the sliding material originated. Over
70% of the sliding material came from the gully banks. Below 2300 m, the solid form of sliding material
is continuous in nature, and the entrainment effects are so significant that boulders (>50 cm in diameter) are
relocated down the slope. The thick mud has a viscosity of about 100 Pa s, and the peak sliding speed reaches as
high as 2 m/s. A total of 2.05 × 106 m3 of solid sliding material was involved in this slide and was spread over an
area of about 3.2 × 106 m2.
Figure 6. A characteristic storm-triggered landslide (a debris flow). This is a plane view of the entire (solid material) collection basin. The elevation divisions are only for reference. The section with concentrated solid material creeping is only a
small portion of the entire area. This means that mass redistribution is referred to as progressive bulking [Iverson, 1997].
REN
©2014. American Geophysical Union. All Rights Reserved.
3649
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Figure 7. The SEGMENT-Landslide simulation of the Sanyanyu gully area. (a) The elevation changes (m) in color, at 9 min
after the 7 August heavy rainfall event. Mud sources are clearly shown in Figure 7a. (b) The final deposition of elevation,
approximately an hour later. The flow has ceased and the deposition is in the Zhouqu city area, via the two-parallel gullies.
Elevation changes of the creeks (the two red arrows) are small and act primarily as pathways for the sliding material. Note
the break/failure of the 3400 m elevation contour, indicating the provision of sliding material for the next sliding cycle.
Figure 7 uses the actual vegetation coverage of the area. If it is assumed that the entire region is vegetation free
(bare ground), SEGMENT-Landslide identifies the following hot spots as unstable (Figure 5b): 33.773°N, 104.375°E;
33.347°N, 104.412°E; 33.77°N, 104.35°E; 33.79°N, 104.38°E; and 33.965°N, 104.105°E. In actuality, only the Sanyanyu
Valley slid significantly. The model shows clearly that it was the light vegetation cover of the Sanyanyu Valley
that was the main reason for such a large-scale outbreak of debris flows. In turn, the light vegetation cover may
have arisen from a positive feedback inherent in successive historical landslide deforestation events [e.g., Bolt
et al., 1975]. Repeated landslides, usually of smaller scale, were investigated in SEGMENT-Landslide simulations
using historical precipitation from the TRMM archive. They show that the rainy seasons of 1998, 2001, 2008, and
2009 all produced landslides capable of destroying the existing vegetation cover. Lighter vegetation cover
lowers the criteria for subsequent landslides. This self-propagating mechanism has no lower limit before
leveling the slope to below the granular material repose angle, at which no further sliding is possible.
In the Zhouqu area, the shear zone depth is variable and depends on the quantity of water penetrating into
the crevasses. For bare ground (e.g., covered by previous landslide deposits or rockfalls caused by historical
earthquakes), runoff readily drains into the crevasses, moistens the granular material, and forms a shear zone
at the bottom (the lowest reaches of the crevasses). Vegetation cover reduces surface runoff through canopy
interception. Roots also assist in the retention of water within the rhizosphere. Thus, with vegetation cover,
runoff water cannot be effectively channeled into the crevasses and much less sliding material will be
involved in the landslide. The cohesion of the granular particles, including loam soil, pebbles from fractured
gray limestone, and sand, is of order 0.1 kPa, far less than the root strength of ~10 kPa.
The presence of aboveground vegetation introduces the following effects: aboveground biomass loading
(gravitational); growing season soil moisture extraction by live roots (hydrological); fortification of the soil
within its extension range (mechanical); a changing soil chemical environment through life processes
(e.g., respiration, absorption of minerals selectively, and secretion of organic substances) and therefore the
bond strength among unit cells (chemical), and wind stress loading (meteorological). The overall effects are the
interaction of the above factors, and it is difficult to generalize before a detailed analysis is carried out that is
specific to a given situation. For example, the fortifying roots have yield strengths larger than dry soils, and the
existence of roots is commonly thought to increase the resistance of soils. For example, it is known that shallow
interlocking root networks can contribute to the mechanical reinforcement of soils [Sidle et al., 1985; Selby, 1993;
Lawrance et al., 1996]. For a pasture species, Selby [1993] estimated the “additional” cohesion as ranging from
REN
©2014. American Geophysical Union. All Rights Reserved.
3650
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
0.1 to 9.8 kPa, with changes in soil moisture. Thus, the mechanical effects of the roots also contribute to slope
stability. However, the presence of roots, especially when there is precipitation, also facilitates water channeling
to deeper depths. After the soil is moistened, the cohesion between the soil and the root surface is reduced
greatly, to negligible strengths <0.001 MPa [Lawrance et al., 1996], and the root strength cannot be effectively
exerted. Also, the effect of roots is to “unify” the soil particles within root distribution range. Once the entire
rhyzosphere soil layer is saturated, the fortifying effects will be totally lost (Figure 9). Thus, a more accurate
summation is that the reinforcement effects from roots are an effective mitigating factor for shallow stormtriggered debris flows.
An additional set of sensitivity experiments was performed to further investigate the role of vegetation in
reducing the magnitude of landslides. If the Sanyanyu Basin had been covered 70% by shrub of negligible
biomass loading, with root strength of 0.1 MPa and coarse root (diameter >1 mm) density of 2 m2, all
residing within the top 2 m of soil, the amount of sliding material available, would be only 1.1 × 106 m3 or
about half the actual volume that was involved in the debris flow of 8 August 2010. If there is a closed cover
(that is, 1.0 vegetation fraction), the sliding material can be further reduced to 104 m3 and primarily involves
only pebbles and protruding boulders at lower elevations. These experiments underline the critical role of
vegetation in reducing the size of progressive bulking storm-triggered landslides.
Importantly, the loss of vegetation did not occur in the 10 years prior to the landslide, and there actually are
clear signs that local vegetation cover has been increasing [Zhao and Running, 2010]. Because the 2008
Wenchuan earthquake deepened the crevasses within the soil mantle and the bedrock, the criteria for stormtriggered landslides became significantly lowered. Large landslides did not occur before 2008 because
previous storms, although equally intense and sometimes with even larger rainfall totals (e.g., 4 September
2001), could not infiltrate into the deeper shear zone without the Wenchuan earthquake’s tearing of the
bedrock. Large landslides did not occur during the 2 years prior to 2010 because in that period, the threshold
precipitation intensity and total were not reached. Those 2 years were relatively dry, as indicated by the total
annual precipitations of 500 mm for 2008 and 480 mm for 2009, as estimated from the TRMM measurements.
The landcover in August 2010 therefore was unable to prevent landslides caused by an intense rainstorm,
largely owing to the legacy of the 2008 Wenchuan earthquake. The sealing of the cracks caused and/or
deepened by the Wenchuan earthquake is a slow process occurring on a time scale of several decades. Thus,
a program of rapid restoration of the vegetation cover over the Zhouqu area is urgently required for
rebuilding that region. Zhouqu County, with an annual precipitation over the last 40 years of only 435 mm/yr,
suggests that the strategic priorities are the restoration of forest on the north facing slope and of a seamless
grass cover for the south facing slopes.
The Sanyanyu deep valley has much coarse granular sliding material, particularly stones and boulders,
because of a self-accumulation mechanism originating from its specific topographical features and because
its loamy soil mantle is more easily dissected by running water. Topographically, the valley has “graded river
beds” because the upper parts (near peaks) are steeper than the lower parts (close to the toes). Thus, the
upper bed slopes are larger than the lower segment’s slope. For lighter precipitation events, the stones and
pebbles cannot roll directly to the toe, stopping at midslope and creating natural barriers to the sliding
material that follows (see the red blobs in Figure 7). These accumulations apply to small slides, typically
caused by low to moderate precipitation events. They have occurred at least 5 times during the past 2 years:
in August 2008, May 2009, June 2009, July 2009, and September 2009. All these SEGMENT-Landslide
simulated events are confirmed by local witnesses. However, when intense precipitation occurs, as in August
2010, all accumulated material will be activated, and a disastrous debris flow will be generated. Studies by Ma
and Qi [1997] and Yu et al. [2010] indicate that granular material accumulated after the June 1880 Wenxian
earthquake [Bolt et al., 1975] was involved as the major debris. This supports the progressive bulking
mechanism. Because previous landslides lacked the lethal combination of extreme rainfall intensity, a prior
large earthquake, and relatively little vegetation coverage, they failed to move the solid material generated
by the Wenxian earthquake down to the Bailongjiang River.
3.2. Future Similar Landslide Events
3.2.1. Precipitation Changes
To estimate possible changes in future extreme precipitation events, WRF was run with thermodynamic
changes computed from the small five-member IPCC GCM ensemble output, under the A1B emission
REN
©2014. American Geophysical Union. All Rights Reserved.
3651
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Figure 8. Schematic 500 mb geopotential heights during longitudinal (high zonal index) blocking in extreme precipitation events for the period 2060–2080. The blue
arrow is the warm conveyor belt; hatched region are surface precipitation areas. Black arrows are cold airstreams from trough pouches. Yellow dashed lines are shear
lines in flow field. Other symbols follow meteorological conventions. Dr. Xinyi Shen of the University of Oklahoma provided the background map of China.
scenario applied to the initial and lateral boundary conditions for this event. The resulting simulation is a
replication of a highly similar synoptic pattern to that of the control simulation, but within a projected
future thermodynamic environment. Because Zhouqu lies in a narrow valley surrounded by high
mountains, there is no low-level jet (LLJ) positive feedback as proposed by Lackmann [2013]. However,
the increased atmospheric water vapor as a direct thermodynamic consequence of warming, or a robust
vapor increase consistent with the Clausius-Clapeyron relationship, still holds [Karl and Knight, 1998;
Semenov and Bengtsson, 2002; Groisman et al., 2005; Emori and Brown, 2005]. In addition, the precipitation
in the Zhouqu area also is strongly influenced by orographic effects. For example, the northern branch of
the circumventing Qinghai-Xizhang Plateau airstream adds cyclonic vorticity to the frontal system. Due
to the complex topography and >3000 m surface elevation, there are no closed surface lows, and there is
only a shear line in the flow fields. The region in question is located within an active cyclogenesis region
to the east of the Qinghai-Xizhang Plateau; there are two such places of frequent cyclogenesis, and
Zhouqu is within the northern one [e.g., Ding, 1991, p. 203]. In this sense, extreme precipitation at
Zhouqu is produced by locally, orographically enhanced convective storms [Li et al., 2012]. In a larger
context, the vapor still eventually comes from the Indian Ocean, following a path illustrated in Figure 8.
Figure 8 is a summary of a series of automatically detected weather patterns from the ensemble Special
Report on Emissions Scenarios (SRES) A1B climate scenario model runs that resulted in extreme
precipitation over the region of interest. The high-low pressure patterns are generic (i.e., they have a
commonly observed structure), and the cases with a nested westward moving tropical cyclone remnants,
REN
©2014. American Geophysical Union. All Rights Reserved.
3652
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Figure 9. Rainfall characteristics and landslide magnitudes. Cyan indicates wet soil and accumulated groundwater. (a) A
vertical cross section of a fractured slope, under dry conditions. Driving stress τ D is exactly balanced by resistive stress τ R
and their magnitudes are smaller than the maximum resistive stress the bedrock can provide τ Dmax, orτ R = τ D < τ Dmax by
several kilopascal. (b) The situation under stratiform precipitation, where rainfall intensity is low but lasts longer. Most
rainfall is uniformly infiltrated to the near surface layer. (c) The situation under convective cloudbursts, which have intense
precipitation that generate runoff and percolation. Most rainfall is either drained into crevasses or simply lost as surface
runoff. In Figure 9b, only the surface layer is mobilized. In Figure 9c, the water discharged to the bottom reduces the
maximum resistive stress significantly. In addition, there is an extra horizontal stress caused by pore water (Ps, pointing
downslope) aiding the driving stress and the entire deep layer is set into motion: τ R = τ Dmax < τ D + Ps. Thus, the same
precipitation amount, with a different morphology, results in different rainfall distributions within slopes and mobilizes
drastically different amounts of sliding material.
which typically make landfall at the Zhejiang coast, are rarer but bringing the largest precipitation totals
over the shaded region on Figure 8. Zhouqu is at the edge of the shaded region and is marginally
affected. In this sense, the climate warming consequence for “outer” circumstances is that of a warm
(and moist) conveyor belt positive feedback like the one proposed by Lackmann [2013]. The enhanced
extreme precipitation in the 2060–2080 time frame may result in larger-scale landslides because of the
following reasoning.
How much sliding material can be mobilized on an unstable slope is codetermined by slope hydrology and
precipitation morphological characteristics (i.e., total amount as well as intensity distribution). For thoroughly
fractured slopes, how much rainfall is infiltrated or drained into the ground is a key parameter for estimating
the landslide magnitude. For example, suppose two rain events have the same precipitation total of 10 mm.
In one case, the rain lasted only 10 min, whereas it lasted 40 min in another case (Figure 9). The dry soil
hydrological properties of a slope are a characteristic quantity of the slope. For example, the infiltration rate is
assumed to be 0.05 m/s. For the case of intense burst that lasted only 10 min, time is insufficient to infiltrate.
The surface runoff, when meets crevasses, percolated into deeper depth, creating a shear zone much deeper
than uniform wetting can reach. The crevasses here can be a result of weathering processes or caused by live
vegetation activities. In cases when the two mechanisms both function, the rainfall percolates even deeper.
REN
©2014. American Geophysical Union. All Rights Reserved.
3653
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
The landslide magnitude thus is much larger for the intense storm case (cf. Figures 9b and 9c). Also, it is
possible that the root reinforcements effects do not work simply because the shear zone becomes far
deeper then the root zone and there is a decoupling of the two. The rainwater may find an indirect route to a
saturated shear zone that does not intersect with root networks. Figure 9 illustrates how rainfall distribution
within the slope affects the landslide magnitude. The weight of rainfall is negligible in aiding the driving
stress. However, if it can be effectively channeled into the interface between nonfractured bedrock and the
overlain granular lump, the effect on reducing the maximum resistive stress is significant. In addition, the
concentrated water column inside the crevasses will exert a hydrostatic horizontal direction differential
pore pressure, aiding the driving stress in overcoming the resistive stress. Once the driving stress and the
differential pore pressure together overtake the reduced (by drainage lubrication) maximum available
resistive stress, the entire layer is set into motion, even though most of the soil inside this layer still is dry. In
contrast, if the precipitation is retained entirely in the upper shallow layer and did not discharge into deeper
layer, only a very shallow layer is mobilized, and this layer, by entraining drier material, quickly loses its
fluidity and may stop midslope. Thus, any future changes in precipitation morphology and structure
characteristics may have direct effects on storm-triggered landslides. This study found that even though
Zhouqu is far inland, the water vapor for precipitating still is nonlocal. As climate warms up, the south-south
west moisture transport increases with low-level jets. However, the precipitating efficiency also is affected
because another factor for frontal precipitation, the provision of cold air, also is adversely affected
(Baroclinic system is slanted westward, with upper level trough to the west of surface low. Precipitation is in
the warm/eastern sector of the surface low. Warm air forced ascending along the upper surface of the cold
air mass is the cause of grid-scale precipitation. The stronger the temperature contrast, the higher the
precipitation efficiency).
There are also examples in which medium followed by severe rainfall intensity maximizes the sliding material
gain. For example, for nonuniform graded slopes, at medium rainfall rate, sliding material cannot slide all the
way to the toe, and it stops the midslope. The following intense rainfall is capable of sending the material to
the toe, entraining along the way at significantly larger rate, because the entrainment process is nonlinearly
related to the instantaneous amount of sliding material. Compared to the washout rate, the accumulation
rate of granular sliding material is much slower. The availability of granular material on the slopes is the first
prerequisite for landslides. Thus, steep slopes are at one end, and the fractured granular mantle is at the
other end.
Compared with the 7–8 August 2010 storm event, future events (cf. Figure 4 and Figure 10a) are expected to
have similar spatial patterns. However, the maximum precipitation amount in the domain increased from
110 mm to 130 mm. The grid point that receives this maximum rainfall amount does not always correspond
to the same geographic location. From Figure 10b, it is also apparent that not all grid points receive more
rainfall as the climate warms, as some locations receive decreased rainfall totals. Examining the regional total
precipitation (Figure 10c for Domain 5, the innermost WRF domain), the precipitation amount indicates that
the total precipitation amounts differ by less than 2% at the end of this rain event. At the time of debris flow
(~18 Z 17 August), the difference in the total amount of rainfall in the simulation domain is not significant.
Zooming into the slope of interest however reveals that the precipitation doubles, reaching ~120 mm in
the first 10 h. Thus, a warmer climate may signify a more spatially concentrated rainfall distribution, with
important landslide consequences.
3.2.2. Bedrock Fracturing Situation
The same precipitation total may mobilize a very different amount of sliding material, as indicated in Figure 9.
The density of macroscopic crevasses (an indicator of the bedrock’s degree of fracture) is critical in the
partitioning of the received precipitation. There are however no direct measurements of bedrock crevasses/
cracks. The remote sensing era provides an opportunity to indirectly deduce the cracking situation of the
bedrocks in the region of interest.
Because of the rainfall’s patchiness, even dense observational stations cannot guarantee the representativeness
of measurements. The annual net primary production (NPP) over a large area however is a good indicator
of annual precipitation [Zhao and Running, 2010]. The area average annual NPP also serves as a smoother/
filter for the spatial rainfall heterogeneity and reflect the changes in area total precipitation accurately.
In the following, the annual area averaged NPP time series are used as surrogates for precipitation
variations over the past decade.
REN
©2014. American Geophysical Union. All Rights Reserved.
3654
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
The Gravity Recovery and Climate
Experiment (GRACE) measurements
(2003 to present) over the periHimalayas show systematic and 100 km
spatial scale consistent signals of mass
loss (blue colors in Figure 11a) and mass
gain (red colors in Figure 11a). First,
the significant mass loss from the
Himalayan region is from the increased
ice mass loss in recent decades [J. Chen
personal communication, 2013;
Wagnon et al., 2007; Jacob et al., 2012].
This meltwater flows primarily into the
rivers and is discharged into the Indian
and Pacific Oceans, with a turnover
time period generally of less than a
week. The three surrounding domains
all have significant mass gains, much
larger than explained by geological
plate movements, which global are
only ~2 GT magnitude, which is 3
orders of magnitudes smaller than the
mass loss. There also have been no
large-scale population and animal
migrations in the past decade. The only
possible cause is in the water/fluids
mass changes. Region 3, which is the
primary agricultural region of China,
with crops comprising the major
vegetation over the region, may exhibit
strong seasonal fluctuations in its
biomass, but annual NPP variations still
are a good representation of the
precipitation amount. Region 2 is an
arid region, which is clearly shown
in the area total biomass (NPP in
Figure 12). Region 2 has a similar total
Figure 10. (a) WRF simulated total precipitation over Domain 5 during 7–8 August
2010 under a warming climate condition
(SRES A1B scenario). (b) The difference
between SRES A1B and current climate
(Figure 4, control). The comparison of
Figure 10a with Figure 4 indicates that the
maximum value has increased from
110 mm to 130 mm. However, Figure 10b
indicates that the increase is not an everywhere unanimous increase (there are
places receiving less rainfall). (c) The time
series of accumulated total precipitation
9 3
amount (in 10 m ) in Domain 5. The
total precipitation differs less than 2%.
Compared with the control case, the distribution is more concentrated.
REN
©2014. American Geophysical Union. All Rights Reserved.
3655
Journal of Geophysical Research: Atmospheres
Figure 11. Mass changes from GRACE over the period 2003–2013 (in
water depth equivalent, cm). For convenience, it is subdivided into four
nested domains (d01-04 or, as defined in the text, Regions 1–4). Regions 2
(23.12–33.54°N, 105–114.5°E), 3 (18.1–24.5°N, 93–106.5°E), and 4 (32.1–
28.5°N, 77–99°E) all show significant mass gain signal, for different physical reasons (discussed in the text).
Figure 12. Area averaged mass changes (GT) relative to 2003 levels, for the
four Regions (also referred to in this study as domains) of interest. A good
indicator of precipitation, NPP, is plotted (black lines) to show possible
trends in precipitation. Only Region 3’s mass changes have strong correlations with precipitation. Regions 2 and 4 both have upward trends unexplainable by precipitation, which has no apparent trends during this period,
as reflected by the NPPs.
REN
©2014. American Geophysical Union. All Rights Reserved.
10.1002/2013JD020881
area as Region 3, but the total NPP is 2
orders of magnitude smaller. Although
the total mass in Region 2 significantly
increased during the 2003–2012
period, the NPP did not. New data from
MODIS indicate a small decreasing
trend. In Region 4, the precipitation
showed a significant decrease during
the past decade (2003–2012), in strong
discord with the mass increase trend.
The precipitation did not increase for
Regions 2 and 4 (it reduced in Region
4); the increased mass can only be
explained by an increased proportion
flowing into groundwater reserves
rather than into runoff (river flow).
Region 2 is less populated and is not a
major agricultural region in China.
Irrigation would only cause mass loss,
which is the opposite of the observed
mass increase, and any attempts to
recharging aquifer are not planned.
This trend possibly can be a result of
increased ground disintegration
(crevasses or cracks in bedrocks). The
reason is not clear but may also have a
natural component after the 2008
Wenchuan earthquake, considering
that the signal of a mass increase starts
at least from 2003, 5 years earlier.
Whatever the causal factors for the
disintegration of the bedrock over
the two domains, there is a reason
to remain on alert for possible
subsequent natural hazards that are
impending. Reiterating the previously
stated general rule: landslides partially
are an upscaling process. Local slope
and geological features enhance its
occurrence in a larger area predisposed
for its development. The growing
bedrock crevasses, and the fluids
accumulated as groundwater, will act
together making future landslides
more severe and more frequent.
Looking further ahead, the pattern
of mass changes creates a saddle
structure in the vicinity of Region 2. The
exerted stress may cause strain buildup
and cause further earthquakes (by
reducing the natural occurrence
period), and this composes a positive
feedback to secondary geohazards
such as flash floods and debris flows.
3656
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Note that the mass gain trend for Regions 2 and 4 began well before the 2008 Wenchuan earthquake and even
before 2003, the starting time of the GRACE measurements. This points to possible geological causes, because
there are no large hydraulic projects that can persistently affect the region of interest.
4. Discussion
The summer storm season of 2010 saw landslides in Zhouqu China (8 August 2010), Sikkim, India (27 August),
and Guatemala (3 September), and flash floods over south central U.S. in (4 May). An obvious question is
whether or not these events are a bellwether of an intensified water cycle, as a consequence of a warming
climate. The modeling study carried out here suggests that the cause of the 8 August debris flows is the result
of an unfortunate combination of multiple factors. The natural events include a heavy rainstorm, not
necessarily the result of global warming, which triggered the landslide and followed a drought that created
surface cracks and crevasses; the geology of the region; and the bedrock damage, which deepened the
surface crevasses, inflicted by the 7.9 magnitude Wenchuan earthquake of 12 May 2008. The human
contribution was historical deforestation and topsoil erosion. Consequently, Zhouqu County became
vulnerable to a devastating rainstorm-triggered landslide. The SEGMENT-Landslide model confirmed the
cause of the landslide by producing a rain-triggered debris flows far larger than historical landslides. The
storm-triggered landslide was magnified by prior vegetation loss and by water penetration deep into
the cracks and crevasses created by the 2008 Wenchuan earthquake. The findings of Ma and Qi [1997] and Yu
et al. [2010], that solid granular material from a historical earthquake in China in 1880, were involved in the
debris flows, further confirm our hypothesis. Note that the rainfall intensity is not of a 100 year recurrence
frequency but only a 20 year recurrence frequency, according to a generalized extreme value analysis.
Consequently, there is a 42% likelihood of a recurrence in the upcoming 10 years. Because the combination
of strong precipitation with poor vegetation and recent earthquake enhancement of the crevasses is lacking
in the past century, previous smaller debris flows failed to transport the granular deposits to stable locations.
It also reflects the difficulty in revegetating the landslide scarps and even the granular deposits for the region,
due to the relatively dry climate.
The massive Zhouqu landslide of August 2010 was caused by an extreme precipitation event, magnified by
the Wenchuan earthquake of May 2008, which greatly deepened the preexisting cracks in the ground
surface, from either an earlier earthquake or more gradual erosion processes. For such cracked surfaces,
intense precipitation events channel runoff water to greater depths than usual, creating sliding surfaces at
those depths. Thus, more sliding material was involved than that typical of a less intense rainstorm. Vegetation
is highly effective at holding drainage water in the rhizosphere and reducing drainage into deeper levels.
However, the severe vegetation loss in the Zhouqu region prevented the vegetation cover from playing such a
protective role in reducing the critical impact of the hydrological process of deep level drainage.
The modeling results underscore the urgency for a high-priority program of revegetation of Zhouqu County,
without which the region will remain exposed to future disastrous, progressive bulking type landslides.
A direct cause of the large magnitude of the 2010 debris flow is the loss of historical deposits and the
undercutting of loose gully bed. Revegetation of the areas with historical deposits therefore is a priority. Thus,
engineering approaches, such as installing check dams, slope protectors, and leveling gullies, should be
followed by revegetation. Restoring the current vegetation cover to its natural, much denser state is the most
effective long-term approach to landslide mitigation. All else being equal, with vegetation present, a
significant portion of rainfall goes into canopy interception (canopy runoff), and runoff and surface ponding
are reduced as a consequence. Vegetation also effectively prevents water infiltrating to greater depths. Thus,
less sliding mass is involved in the vegetated case.
In the global water cycle, precipitation over land generally exceeds evapotranspiration, the excess water
usually as river runoff getting back to oceans to close the cycle (over oceans, evaporation exceeds
precipitation). Hence, the water vapor condensed in extreme precipitation events is not local to the
precipitation but rather it is advected from remote source regions by an atmospheric counterpart of the
surface river: the LLJs [Harrold, 1973; Browning and Pardoe, 1973]. The LLJ is a synoptic scale phenomenon
(~1000 km long and 100 km wide airstream), usually located to the southeast of the cold front in a
midlatitude frontal system that brings in a warm and moist air mass to the precipitation region. To make
the situation persistent, i.e., slowing down the system evolution, a high-latitude blocking configuration also
is needed (Figure 8). The blocking system was responsible for the antecedent significant precipitation events
REN
©2014. American Geophysical Union. All Rights Reserved.
3657
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
(e.g., 21 and 24 July and 4 August 2010) over Zhouqu before the decisive 7–8 August event. With an
enhanced hydrological cycle [Trenberth, 1999], more extreme weather conditions are expected in the future.
For the Zhouqu area, in addition to the mechanism reported by Lackmann [2013], orographic precipitation is
also a factor. The northern branch of the circumventing Qinghai-Xizhang Plateau airstream adds cyclonic
vorticity to the frontal system. The region in question is located within an active baroclinic cyclogenesis area
to the east of the Qinghai-Xizhang Plateau [see e.g., Ding, 1991, p. 203]. The precipitation has a 20 year
recurrence frequency, as calculated from projected climate change. A disaster like 2010 is expected within
~20 years if no effective countermeasures are taken. In addition, the large environment around this region
clearly still is favorable for future severe debris flows.
Remotely sensed observations over the past decade provide an opportunity to examine the bedrock
cracking conditions over the peripheral Himalayan region, which includes both Zhouqu County and the areas
affected by the 2008 Wenchuan earthquake. There are two regions experiencing steady mass gains over the
past decade. It is deduced that the mass increase is likely due to increased percolation of surface runoff, or
recharging of the fossil water (a groundwater reservoir), rather than an increase in precipitation. The reason
for the increased partitioning of precipitation into groundwater is not entirely clear, but an increase in
bedrock crevasses is one viable explanation. To worsen the situation, these two regions of ever-increasing
fracturing are colocated with the atmospheric baroclinic cyclogenesis basins corresponding to the QinghaiXizhang Plateau airstreams. As the climate warms, extreme precipitation events are expected to become
more common [e.g., Trenberth, 1999; IPCC, AR4, 2007], so future recurrence frequencies of such events
therefore must be estimated and included in the forcing of the landslide model.
Thus, landslides hold a unique nexus location in geomorphology research (joints the tectonic-climatetopography-erosion-river incision interlinks), unifying several disciplines in subjects ranging over many
different spatial and temporal scales. Because complicated landforms are the products of deep earth
processes involving mantle convection and plumes producing plate tectonics, orogeny and epeirogeny, as
well as the surface processes of soil formation, denudation, and sedimentation, influenced by seismicity,
climate change, and land cover change, with a recent (on geological time scales) significant role for a
human contribution. In addition, processes and rates of change vary with spatial and temporal scale in an
interconnected hierarchy. Hence, the land surface become so complicated that there is no simple way to
scale up—processes that operate at small spatial scale will not provide either an adequate description or
understanding of landforms at larger spatial scales. To account for the stability of slopes, only a full physics
numerical model such as that employed in this study, SEGMENT-Landslide, a surface transport type of model
in geomorphology term, is not adequate. Knowledge of the slope material properties over a large enough
neighboring region also is a prerequisite.
Appendix A: The SEGMENT-Landslide Model
A landslide is a “fatigue” symptom of granular slopes. The general theories of material properties apply but
require modifications. The occurrence of landslides can be analyzed from an energy point of view (the
maximum tolerable distortional energy). In SEGMENT-Landslide, the full internal stress tensor is bundled as
σ ¼ ϕ þ C þ Sp μ þ δE ;
(A1)
where ϕ = ρgh is the gravitational potential, C is the effective cohesion (also called “apparent cohesion” or
“interlocking cohesion,” i.e., the shear stress which the material can sustain at zero normal stress), and δE is the
pressure perturbation caused by earthquake or human-induced disturbances at that location. The
μ < μ1 = tgф, with ф granular repose angle (i.e., angle of friction). It is found experimentally that the behavior
of soil materials above the water table can be adequately represented by the Coulomb-Navier fracture
criterion (the middle two terms of equation (A9)). The Coulomb-Navier criterion is a special case of Mohr
theory with the straight line, tangential to the circles of tension-simple shear-compression, the Mohr
envelope. For conditions with groundwater, hydrostatic pressure (pore pressure) is usually included in Sp for
convenience. Except around cavities, where the so-called “bridge effects” distort the stress fields, Sp is close to
the loading pressure. The extreme values of the middle two terms on the right-hand side are the yielding
strength (shear strength) of the sliding material τ f = C + Sp,f μ, with subscript “f” meaning failure. This is the
Navier-modified Columb criterion. C and μ are functions of soil moisture, soil chemical components, and also of
shear stress [e.g., Schofield and Telford, 2006]. For soils at depths greater than 10 m, the dependence of μ on
REN
©2014. American Geophysical Union. All Rights Reserved.
3658
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
stress is minimal, and the Navier-modified Columb criterion is more applicable than for shallower soils (i.e.,
smaller Sp, Kenney [1984]; Townsend and Gilbert [1973]). When applied to shallow shear surfaces, volume
change and dilation within the shearing soil have an apparent nonlinear influence on μ. For sandy soils, the
friction angle increases with increasing confining pressure [Cornforth, 2005; Lu and Godt, 2008]. For an
extremely deep shearing zone (>200 m), von Mise’s criterion is a closer approximation, because the plasticity
of earth material becomes more salient.
For nonfractured bedrock, C is the dominant term and is usually 3 orders of magnitude larger than the
remaining three terms combined. For most of the soil (except pure sandy soil), cohesion and internal friction
are both important in maintaining stable slopes. For fractured rocks and sandy soils, the internal friction
becomes the dominant term, although it is not necessarily larger than the gravitational potential. This is
because the horizontal gradient of the gravitational potential causes motion instead of the bulk term. For
natural slope soil, both the cohesion and friction components of the strength are sensitive to the clay particle
content. On the one hand, the large surface area provided by the clay minerals enhances the development of
cohesion. On the other hand, the interlocking friction is weakened. In shear box tests, interlocking friction is a
measure of the amount of deviation required for particles at the shear surface to move past each other. A soil
dominated by clay particles requires very little deviation to be sheared. In northwest Europe, there is a type of
“quick” clay. The soluble ingredients in the structure of a solid make its rheological properties even more
sensitive to moisture contents. For vegetated surfaces, C also includes the fortification from distributed roots.
The SEGMENT-landslide model is extensively documented by Ren et al. [2008, 2011]. Here governing
equations are presented to provide a context for this study. For the sliding material, a coupled system is
solved for conservation of mass:
→
∇ V ¼ 0;
(A2)
and for conservation of momentum:
8
∂σxy
∂σxx
∂σxz
du
>
>
þ
þ
¼ ρ
>
>
dt
∂x
∂y
∂z
>
>
>
< ∂σ
∂σyy
∂σyz
dv
xy
þ
þ
¼ ρ
> ∂x
dt
∂y
∂z
>
>
>
>
∂σ
∂σ
∂σ
dw
>
yz
xz
zz
>
:
þ
þ
ρg ¼ ρ
dt
∂x
∂y
∂z
;
(A3)
under the granular rheological relationship, with viscosity parameterized as
v¼
μ0 þ
μ1 μ0 S
;
I0 =I þ 1 jε• ej
(A4)
→
where ρ is the bulk density, V is the velocity vector (u,v, and w are the three components), σ is the internal
stress tensor, and g is the gravity acceleration. Failure of sloping earth materials falls within the general form
of fatigue. However, the involved material is not a single solid specimen, rather, a multiple body of countless
degree of freedom is involved: a lump of granular material, with rocks at the coarse end of grain size and
sandy loam soil at the fine end of grain size. Parameterizing the granular viscosity is critical for a dynamic
landslide model. In equation (A4), ν is the viscosity, S is the spherical part of the stress tensor σ, μ0 and μ1 are
•
•
• • 0:5
the limiting values for the friction coefficient μ, jεej is the effective strain rate and jεej ¼ 0:5 εij εij , I0 is a
constant depending on the local slope of the footing bed as well as the material properties, and I is inertial
•
number defined as I ¼ jεejd=ðS=ρs Þ0:5, where d is particle diameter and ρs is the particle density. Soil moisture
enhancement factor for viscosity is assumed to vary according a sigmoid curve formally as equation (A9) of
Sidle [1992] but with the time decay term replaced by relative saturation.
Under the unique processes of earth environments, the mechanical strength of the granular material is
further altered by the presence of vegetation roots (fortification effects) or tunnels of subterranean animals
(weakening effects). For granular material resting on vegetated slopes, the cohesion provided by the roots is
implemented in the full internal stresses σ ’ s. The root mechanical properties are prescribed according to the
vegetation types of the Zhouqu area. There are different ways to decompose the full stress into spherical and
deviatoric components. Only the deviatoric part is assumed to be proportional to the strain rate through
viscosity, which also depends on normal pressure.
REN
©2014. American Geophysical Union. All Rights Reserved.
3659
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
As a derivative from equation (A2), the prognostic equation for surface elevation (h(x,y)) is
∂h → þ V•∇H h wjtop ¼ 0;
top
∂t
(A5)
where X|top indicates the evaluation at the free surface elevation. In the case with slope movements, equation (A5)
is solved regularly to update the sliding material geometry. It is also from this equation that we estimated the
sliding material involved in the simulated landslides (e.g., Figure 6).
The viscous term in equation (A3) implies an energy conversion from kinetic energy to heat. To make a full
closure of energy, we need the following thermal equation:
∂T → 2
ρc
þ V ∇ T ¼ kΔT þ σ2e ;
(A6)
∂t
ν
where c is the heat capacity (J/kg/K), T is temperature (K), κ is the thermal conductivity (W/K/m), and σe is the
effective stress (Pa). The last term is “strain heating,” converting of work done by gravity into heat affecting
the sliding material by changing viscosity or causing a phase change.
A further level of complexity that acts on slope stability within the ecosystem of a sloping terrain is the effect
on the granular material’s moisture content, through slope hydrology. SEGMENT-Landslide solves the full
form of Richards equation to obtain the slope moisture conditions. Fluid flow and moisture content in hilly
environments vary spatially and temporally, due to time-dependent environmental changes (e.g.,
precipitation, solar radiation, and transpiration of vegetation through distributed roots) and the storage
capacity of porous granular soils. In modeling, the time-dependent environmental changes are usually
accounted for by the boundary conditions of distributed forcing terms, whereas the soil storage is usually cast
in the flow laws. Considering a representative soil volume (RSV), this RSV is of arbitrary volume and arbitrary
→
surface area Ω with an outward pointing normal unit vector n. We assume that the soil porosity is P
(maximum of the volumetric water content w ) and moisture flux is→
q.
c
From Green’s first theorem and mass continuity,
→ →
→ ∂ρ w c
∇ • ρw q ¼ q • Ω n ¼ w :
∂t
Equation (A7) holds everywhere inside the porous media. From Darcy’s law, soil water flux (kg/s) is
proportional to hydraulic conductivity (K) and total hydraulic gradient (ht):
→
q ¼ K∇h
(A7)
(A8)
t
where ht usually is assumed to be the sum total of matric potential (hm, head due to pore water pressure),
head due to gravity (hg), head due to osmosis (ho), and head due to kinetic energy (hv, negligible). Darcy first
realized that the total energy of the system stored in the form of the total water potential is always lost during
flow through the porous material and the amount of energy loss depends on the material. This is a
manifestation of the second law of thermodynamics. In general, energy stored or released by an element of
soil due to change in the fluid mass can be related to the hydraulic head [Freeze and Cherry, 1979, p. 51]:
∂ρw w c
∂hm
¼ ρw 2 gðαs þ Pαw Þ
(A9)
∂t
∂t
where αs is the compressibility of bulk soil and αw is compressibility of water voids. The (αs + Pαw) term is also
the mean compressibility of wet soil (1/Pa). Whereas the compressibility of water under earth surface
environment is close to a constant value of about 4.4 × 1010(1/Pa), the compressibility of clay soil ranges
from 106 to 108, for sandy soil from 107 to 109, and 108 to 1010 for pebbles/gravel. Assuming the
constant water density in equation (A7), and substituting in equations (A8) and (A9), gives
∂hm
(A10)
∂t
Equation (A10) is applicable to saturated and unsaturated flow and is the most general form of Richards
equation. For unsaturated situations, the compressibility of the bulk soil usually is smaller than the
compressibility of the fluid voids, so the void compressibility dominates the average compressibility.
By definition
∇ • ðK∇ht Þ ¼ ρw gðαs þ Pαw Þ
Pαw ¼
REN
P
∂V w
:
V v ∂ðρw ghm Þ
©2014. American Geophysical Union. All Rights Reserved.
(A11)
3660
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
Substituting equation (A11) into equation (A10) and rearranging give
∇ • ðK∇ht Þ ¼
∂w c
:
∂t
(A12)
In the absence of osmotic head, the total head, comprising only matric suction head and geopotential, gives
the most common form of Richards equation (3-D hydraulic conductivity in the component form):
∂
∂hm
∂
∂hm
∂
∂hm
∂K z ∂w c
Kx
Ky
Kz
þ
þ
þ
¼
(A13)
∂x
∂y
∂z
∂x
∂y
∂z
∂z
∂t
For 3-D landslide models, the land surface model subcomponent must solve Richards equation in a full 3-D
setting (equation (A13)). For example, SEGMENT-Landslide includes SHEELS [Ren et al., 2004] as a subscheme
to simulate the soil moisture content. For the application to 2-D (x-z plane) slope models (e.g., TRIGRS [Baum
et al., 2008]), Richards equation in hill slope settings is proposed based on Philip’s [1991] assumption,
although several recent theoretical and experimental studies question the validity of this assumption [e.g.,
Jackson, 1992; Philip, 1993]. The following two analytical solutions to the Richards equation are valuable in
dealing with flux boundary conditions. The Green and Ampt [1911] infiltration model is an analytical solution
for flow into an initially dry, uniform column of soil. Although not a direct analytical solution of the Richards
equation, it has the essence by two assumptions arising from observed wetting front during infiltration: (A1)
the soil suction beyond (ahead of) the wetting front is constant in space and time and (A2) the water content
and the corresponding hydraulic conductivity of the soil behind the wetting front (the wet portion of soil
column) also are constant in space and time. For some efficient land surface schemes, this relation is utilized
for unsteady rainfall flux upper boundary conditions [e.g., Chu, 1978]. The infiltration scheme inside TRIGRS is
also a unique analytical solution of the equation (A13). It is based on the Gardner’s [1958] linearization of a 1-D
Richards equation applied to a water table. An analytical solution for transient infiltration above the
groundwater table is thus obtained [Srivastava and Yeh, 1991; Baum et al., 2008]. Suction head and soil
moisture profiles, as functions of vertical elevation above water table and of time, are standard outputs of this
type of model.
For quantitative predictions of storm-triggered landslides, a numerical modeling system like SEGMENTLandslide is required. However, some of the requirements of SEGMENT, especially the input and verification
data, generally are not available even in modern geological maps/archives. These parameters include
vegetation loading and root distributions in soils and weathered rocks. The extension of the SEGMENT
landslide model to other regions is limited primarily by the lack of these high-resolution input data sets. The
landslide features implemented in SEGMENT-Landslide, if adopted by the relevant community, hopefully will
encourage the collection of such vital information in future surveys.
Acknowledgments
We thank Maosheng Zhao and
Xiaoyang Zhang for their useful discussions on MODIS remote sensing
biomass data and the allometric
approaches in determining the total
biomass from MODIS data. Chris Harris
assisted with utilizing the supercomputing facility at Curtin University.
REN
References
Balkema, A., and L. de Haan (1974), Residual lifetime at great age, Ann. Probab., 2, 792–804.
Baum, R., W. Savage, and J. Godt (2008), TRIGRS-A Fortran program for transient rainfall infiltration and grid-based regional slope-stability
analysis, version 2.0. US Geological Survey Open-File Report, 75pp.
Bolt, B., W. Horn, G. Macdonald, and R. Scott (1975), Geological Hazards., 328 pp., Springer-Verlag, New York.
Browning, K., and C. Pardoe (1973), Structure of low-level jet streams ahead of mid-latitude cold fronts, Q. J. R. Meteorol. Soc., 99,
619–638.
Chu, S. (1978), Infiltration during an unsteady rain, Water Resour. Res., 14, 461–466.
Coles, S. (2001), An Introduction to Statistical Modeling of Extreme Values, Springer Series in Statistics, 224 pp., Springer Verlag, London.
Cornforth, D. (2005), Landslides in Practice: Invstigation, Analysis, Remedial/Preventative Options in Soils, John Wiley, New York.
Ding, Y. (1991), Advanced Synoptic Meteorology, 792 pp., China Meteorologocal Press, Beijing.
Emori, S., and S. Brown (2005), Dynamic and thermodynamic changes in mean and extreme precipitation under changed climate, Geophys.
Res. Lett., 32, L17706, doi:10.1029/2005GL023272.
Freeze, R., and J. Cherry (1979), Groundwater, Prentice Hall, Englewood Cliffs, N. J.
Gardner, W. (1958), Some steady-state solutions of unsaturated moisture flow equation with application to evaporation from a water table,
Soil Sci., 85, 228–232.
Green, W., and G. Ampt (1911), Studies on soil physics, Part I. The flow of air and water through soils, J. Agr. Sci., 4, 1–23.
Groisman, P. Y., R. W. Knight, D. R. Easterling, T. R. Karl, G. C. Hegerl, and V. N. Razuvaev (2005), Trends in intense precipitation in the climate
record, J. Clim., 18, 1326–1350.
Harrold, T. (1973), Mechanisms influencing the distribution of precipitation within baroclinic disturbances, Q. J. R. Meteorol. Soc., 99,
232–251.
IPCC, AR4 (2007), Climate Change 2007. The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change, edited by S. Solomon et al., Cambridge Univ. Press, Cambridge, U. K. and New York.
©2014. American Geophysical Union. All Rights Reserved.
3661
Journal of Geophysical Research: Atmospheres
10.1002/2013JD020881
IPCC, AR5 (2013), Climate Change 2013. The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change, edited by T. Stocker et al., Cambridge Univ. Press, Cambridge, U. K. and New York. [Available at
https://www.ipcc-wg1.unibe.ch/.]
Iverson, R. (1997), The physics of debris flows, Rev. Geophys., 35, 245–296.
Jackson, C. R. (1992), Hillslope infiltration and lateral downslope unsaturated flow, Water Resour. Res., 28, 2533–2539.
Jacob, T., J. Wahr, W. Pfeffer, and S. Swenson (2012), Recent contributions of glaciers and ice caps to sea level rise, Nature, 8, doi:10.1038/
nature10847.
Karl, T., and R. Knight (1998), Secular trends of precipitation amount, frequency, and intensity in the USA, Bull. Am. Meteorol. Soc., 79, 231–241.
Kenney, C. (1984), Properties and behaviours of soils relevant to slope instability, in Slope Instability, edited by D. Brunsden and D. B. Prior,
pp. 27–65, John Wiley, Chichester.
Lackmann, G. (2013), The south-central US flood of May 2010, present and future, J. Clim., 26, 4688–4709.
Lawrance, C., R. Rickson, and J. Clark (1996), The effect of grass roots on the shear strength of colluvial soils in Nepal, in Advances in Hillslope
Processes, edited by M. G. Anderson and S. M. Brooks, pp. 857–867, John Wiley, Chichester.
Li, J., C.-Y. Liu, P. Zhang, and T. J. Schmit (2012), Applications of full spatial resolution space-based advanced infrared soundings in the
preconvection environment, Weather Forecasting, 27, 515–524.
Lu, N., and J. Godt (2008), Infinite slope stability under unsaturated seepage conditions, Water Resour. Res., 44, W11404, doi:10.1029/
2008WR006976.
Ma, D., and L. Qi (1997), Study on comprehensive controlling of debris flow hazards in Sanyanyu Gully, Bull. Soil Water Conserv., 17, 26–31.
Philip, J. (1991), Hillslope infiltration: Planar slopes, Water Resour. Res., 27, 109–117.
Philip, J. (1993), Comment on “Hillslope infiltration and lateral downslope unsaturated flow” by C R. Jackson, Water Resour. Res., 29, 4167.
Pielke, R. A., Sr., R. Wilby, D. Niyogi, F. Hossain, K. Dairuku, J. Adegoke, G. Kallos, T. Seastedt, and K. Suding (2012), Dealing with complexity and
extreme events using a bottom-up, resource-based vulnerability perspective, in Extreme Events and Natural Hazards: The Compexity
Perspective, Geophys. Monogr. Ser., vol. 196, edited by A. S. Sharma, pp. 345–359, AGU, Washington, D. C.
Ren, D., M. Xue, and A. Henderson-Sellers (2004), The effects of hydraulic lift in simulating superficial soil moisture for vegetated surfaces
under dry conditions, J. Hydrometeorol., 5, 1181–1191.
Ren, D., L. Leslie, and D. Karoly (2008), Mudslide risk analysis using a new constitutive relationship for granular flow, Earth Interact., 12, 1–16.
Ren, D., J. Wang, R. Fu, D. Karoly, Y. Hong, L. Leslie, C. Fu, and G. Huang (2009), Mudslide caused ecosystem degradation following the
Wenchuan earthquake 2008, Geophys. Res. Lett., 36, L05401, doi:10.1029/2008GL036702.
Ren, D., R. Fu, L. M. Leslie, and R. Dickinson (2011), Predicting storm-triggered landslides, Bull. Am. Meteorol. Soc., 129–139,
doi:10.1175/2010BAMS3017.1.
Schofield, A., and T. Telford (2006), Disturbed soil properties and geotechnical design, ISBN 0-7277-2982-9.
Selby, M. (1993), Hillslope Materials and Processes, 2nd ed., 451 pp., Oxford Univ. Press, New York.
Semenov, V., and L. Bengtsson (2002), Secular trends in daily precipitation characteristic greenhouse gas simulation with a coupled AOGCM,
Clim. Dyn., 19, 123–140.
Sidle, R. (1992), A theoretical model of the effects of timber harvesting on slope stability, Water Resour. Res., 28, 1897–1910.
Sidle, R., A. Pearce, and C. O’Loughlin (1985), Hillslope Stability and Land Use, AGU, Washington, D. C.
Skamarock, W. C., et al. (2008), A description of the Advanced Research WRF version 3. Tech. Rep. NCAR/ TN-4751STR, 113 pp.
Srivastava, R., and T.-C. Yeh (1991), Analytical solutions for one-dimensional, transient infiltration toward the water table in homogeneous
and layered soils, Water Resour. Res., 27, 753–762.
Townsend, F., and P. Gilbert (1973), Tests to measure residual strengths of some clay shales, GeÂotechnique, 23, 267–271.
Trenberth, K. (1999), Conceptual framework for changes of extremes of the hydrological cycle with climate change, Clim. Change, 42,
327–339.
Wagnon, P., et al. (2007), Four years of mass balance on Chhota Shigri Glacier, Himachal Pradesh, India, a new benchmark glacier in the
western Himalaya, J. Glaciol., 53, 603–611.
Yu, B., Y. Yang, and Y. Su (2010), Research on the giant debris flow hazards in Zhouqu county, Gansu province on August 7, 2010 [in Chinese],
J. Eng. Geol., 18, 437–444.
Zhang, X., and S. Kondragunta (2006), Estimating forest biomass in the USA using generalized allometric models and MODIS land products,
Geophys. Res. Lett., 33, L0940, doi:10.1029/2006GL025879.
Zhao, M., and S. Running (2010), Drought-induced reduction in global terrestrial net primary production from 2000 through 2009, Science,
329, 940–943.
REN
©2014. American Geophysical Union. All Rights Reserved.
3662