Rahardjo et al.

Urban Lifeline (2023) 1:5

https://doi.org/10.1007/s44285-023-00006-9
Urban Lifeline

RESEARCH Open Access

Slope susceptibility map for preventive

measures against rainfall‑induced slope failure
Harianto Rahardjo1* , Qian Zhai2, Alfrendo Satyanaga3, Yangyang Li4, Saranya Rangarajan1 and Arezoo Rahimi5

Abstract
Rainfall-induced slope failures frequently occurs in many urban areas around the world especially due to the impact
of the global warming. Some slope failures result in casualties that have negative impacts on urban sustainability.
The slope failures are normally observed near slope surface within the unsaturated zone above the groundwater
table. Hence, it is important to incorporate the unsaturated soil mechanics principles in analysing the slope stability
during rainfall. This study focuses predominantly on the regional stability analyses of residual soils within the Jurong
Formation and Bukit Timah Granite in Singapore. The objective of the study is to develop slope susceptibility map
as part of preventive measures against extreme rainfall conditions. The slope susceptibility map was evaluated
using 2-D numerical analyses of slopes with historical failure. The recently developed sustainable retaining structure
system, GeoBarrier System (GBS) is investigated to understand its suitability as a preventive measure for critical slopes
with high risk as identified from the slope susceptibility map. Using Transient Rainfall Infiltration (TRIGRS) and Grid-
Based Regional Slope Stability Analysis (Scoops3D), the pore-water pressure distributions and factors of safety (FS)
within each zone are determined. The results from slope susceptibility maps and 2-D numerical analyses are consist-
ent. Moreover, the incorporation of GeoBarrier System on the critical slope is demonstrated to be an effective slope
stabilization measure based on the numerical simulations of two-dimensional analyses.
Keywords Urban sustainability, Slope susceptibility map, Unsaturated soil mechanics, GeoBarrier System, Rainfall-
induced slope failure

1 Introduction
Slope failures are a common phenomenon that can lead
to significant losses. In 1988, a massive landslide in Dar-
bang, located approximately 200 km west of Kathmandu,
killed 109 people and temporarily blocked the Myagdi
River. Furthermore, years prior to that event, a landslide
*Correspondence: buried the Darbang region and caused the death of 500
Harianto Rahardjo
people [63]. Landslides triggered by rainfall also occurred
[email protected]
1
School of Civil and Environmental Engineering, Nanyang Technological in China, as well as in other countries with extreme
University, Block N1, 50 Nanyang Avenue, Singapore, Singapore geological catastrophes. In China, geological disasters,
2
Bridge Engineering Research Center, Southeast University,
including landslides, account for more than 20 percent
Nanjing 210096, China
3
School of Engineering and Digital Sciences, Nazarbayev University, of the total economic losses caused by natural disas-
Kabanbay Batyr Ave. 53, Astana 010000, Kazakhstan
4
ters each year. The nationwide direct and indirect eco-
Monash Suzhou Research Institute, Monash University, Suzhou Industrial
Park, Suzhou, China
nomic losses related to landslides exceed 20 billion Yuan
5
Faculty of Engineering, Civil and Environmental Engineering, (approximately 2 billion EUR) annually [1, 21]. In Japan,
The University of Auckland, Engineering Block 1, 20 Symonds St, heavy rainfall has led to recurring landslides over the past
Auckland 1010, New Zealand
65 years, resulting in more than a thousand deaths [6].

© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 2 of 21

Every year, thousands of lives and billions of dollars probability of a landslide based on data about the variables
are wasted due to landslides, as reported by [10] Far- associated with its occurrence [16]. These models assume
ahmand and Aghakouchak. The 1999 landslide in Var- that there are established relationships between landslide
gas, Venezuela [30], for example, was one of several susceptibility and the variables used for assessment [57].
landslide disasters that killed and displaced thousands This technique involves using a detailed geomorphological
of people. In South and Southeast Asia, landslides are map and a wide range of factors as an attribute database to
one of the most prevalent disasters due to the region’s quickly assess landslide susceptibility following fieldwork
weather, topography, and socioeconomic conditions, [22]. Nevertheless, this method is limited by the repro-
landslides are one of the most prevalent disasters in ducibility of its outcomes and the subjectivity involved in
South and Southeast Asia [4]. For instance, in 2006, a assigning weightings and ratings to the variables.
series of landslides on Leyte Island, Philippines, killed The first step in any statistical analysis of landslides is
over a thousand people due to extreme rains [55], to create a detailed map of past landslide occurrences
and on October 4, 2010, 145 people were killed by and collect data on all the relevant factors. The occur-
the Wasior landslide in West Papua, Indonesia. The rence of landslides can be predicted using statistical
effects of precipitation on residual soil slopes in Sin- models, which include identifying the permutations
gapore were investigated by Fredlund and Rahardjo of factors that have contributed to such events in the
[11], Krisnanto et al. [28], and Rahardjo and Satyanaga past. Researchers have studied a wide range of statisti-
[45] using tensiometers and piezometers. In Singapore, cal methods, but they always start with the same regular
slope failures are primarily triggered by intense rain- grid cells as their fundamental analytical unit. Bivariate
falls where residual soils are prevalent [47, 49]. Resid- statistics have emerged as an alternative to more com-
ual soil is typically found in the unsaturated zone above plex multivariate statistics, wherein each explanatory
the water table where the negative pore-water pressure variable is represented as a distinct thematic data layer
(i.e., matric suction) in the unsaturated soil contributes in a Geographic Information System (GIS) [25]. Among
to its shear strength [11]. When water infiltrates the these methods, the rainfall threshold approach is often
slope, the pore-water pressure increases and the addi- considered the most effective. Rainfall threshold refers
tional shear strength provided by the matric suction to the minimum intensity or duration of rainfall required
decreases, thereby increasing the susceptibility of the to trigger a landslide, and the commonly used param-
slope to failure [58]. eters for designing rainfall thresholds include cumulative
Due to its tropical climate, Singapore experiences high rainfall, antecedent rainfall, rainfall intensity, and rainfall
levels of rainfall and temperatures that promote rapid duration [18].
in-situ chemical and mechanical weathering. This leads The physical processes that led up to the landslide
to the formation of thick residual soil profiles [19]. Slope occurrence are described in a deterministic analysis (or
instabilities often occurred in residual soil slopes due to physical-based models) for assessing risk. Slope stability
the local climate and geology. To prevent damage to sur- assessments can be used as the deterministic analysis,
rounding buildings or public amenities, it is necessary to and the transient groundwater response of slopes to rain-
protect slopes from the danger of rainfall-induced slope fall can be taken into account. Long-term landslide data
failures [17]. Developing a slope susceptibility map to iden- are unnecessary for deterministic analysis. Therefore, this
tify areas at high risk for slope failure is one of the most method may be used in places where landslide data are
sustainable approaches for prevention of slope failures lacking. Three criteria, including assessment goals, study
[56]. There exist various methods to evaluate the suscep- area size, and data availability, can influence the scale
tibility of slopes, which can be grouped into four main selected for slope susceptibility maps [42].
categories: inventory, heuristic, statistical, and determin- Many researchers have developed physical models to
istic [8, 22]. Inventory analysis, also known as distribu- assess regional slope instability [2, 23, 34, 39]. These mod-
tion analysis, is a straightforward qualitative approach els incorporate topography, saturated soil properties, and
employed to map landslide susceptibility [36]. It involves hydrological parameters. However, they often overlook
depicting landslide details, such as their spatial and tem- the spatial distribution of the groundwater table in their
poral patterns, movement types, rates, and the materials calculations. Their models make use of both high-qual-
displaced (e.g., soil, debris, or rock) [7]. Inventory maps ity data and varying degrees of complexity and difficulty.
serve as simplified forms of susceptibility maps by indicat- Previous studies on slope susceptibility mapping have
ing the locations of documented landslides. However, this relied on physically based models that integrate hydro-
method does not identify areas that might be suscepti- logical characteristics to calculate pore-water pressure
ble to future landslides unless previous occurrences have distributions and use infinite slope analysis to determine
been recorded. Heuristic analysis is utilized to estimate the the factor of safety. Notable examples of such models
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 3 of 21

include SHALSTAB [38], SINMAP [43], GEOtop-FS and and 2-D seepage and stability assessments, as well as sat-
SLIP [39]. These models follow a similar concept known urated and unsaturated laboratory testing.
as the hydrological-infinite slope stability model. TRI-
GRS is the only program that incorporates unsaturated 2 Site overview and climatic conditions
principles to establish the failure criterion [2]. For simu- The geological characteristics of Malaysia and Singapore
lating groundwater flow over time and determining the exhibit remarkable similarities. In the region of Singa-
appropriate safety factor during wet periods, YS-Slope pore, which is situated at the southern tip of the Malay-
can be employed [24]. sian Peninsula, the primary parallel belts extending from
Once areas of concern are identified through slope the north-northwest to the south-southeast are clearly
susceptibility mapping, it becomes crucial to implement defined [35]. Singapore Island is positioned between
effective slope protection measures. Traditional protec- Malaysia’s Eastern Belt and Central Belt due to this geo-
tion systems often have drawbacks such as high costs, logical arrangement. Within the Eastern Belt, Permian
lengthy construction time, and aesthetic incompat- conglomerates dominate the geological formations, with
ibility. However, advancements in slope protection tech- metamorphosed Carboniferous sandstones and mud-
niques, such as the GeoBarrier System (GBS) developed stones (approximately 350–300 million years old) overly-
by Rahardjo et al. [51], offers promising alternatives. ing them. In addition to granites and volcanic rocks from
The GBS utilizes a capillary barrier system as a retain- the Middle Triassic period, the Eastern Belt also features
ing structure, incorporating geosynthetic geobags filled Late Permian shallow marine deposits. The Central Belt
with soil or granular materials. This innovative approach was formed between 275 and 240 million years ago (Mid-
not only provides stability but also aligns with sustain- dle Permian and Middle Triassic) from mudstone and
able practices by utilizing recycled materials like Recy- limestone layers. These strata are interspersed with igne-
cled Concrete Aggregate (RCA) and Reclaimed Asphalt ous rocks like granite and andesite. The Late Triassic
Pavement (RAP) in the capillary barrier system [50]. By sandstones and conglomerates that make up the middle
integrating the capillary barrier system, the GBS ensures belt are what set it apart from the eastern belt.
effective slope protection while mitigating the drawbacks Singapore’s geological landscape is characterized by
associated with traditional methods. In the capillary bar- four dominant formations: 1) Bukit Timah Granite, 2)
rier system, the fine-grained recycled materials (fine RCA Jurong Formation, 3) Old Alluvium, and 4) Kallang For-
or RAP) are placed on top of the coarse-grained recycled mation. The Bukit Timah Granite Formation primar-
materials (coarse RCA or RAP) [60]. The placement of ily comprises Bukit Timah granite and Gombak norite,
bags between the fine- and coarse-grained materials does which are igneous rocks. This formation is predominantly
not compromise the integrity of the capillary barrier sys- found in the northern and central-northern regions of
tem, as demonstrated by Rahardjo et al. [50]. Addition- Singapore. The Jurong Formation, on the other hand, is a
ally, bags filled with an approved soil mixture (ASM) can sedimentary geological formation located in the western
be positioned in front of the fine-grained layer to accom- and southwestern regions. The eastern region of Singa-
modate the future planting of trees and shrubs with deep, pore is characterized by the presence of Old Alluvium, a
extensive taproots. This allows for the integration of quaternary deposit depicted in Fig. 1. Lastly, the Kallang
greenery into the system. Formation is the most recent alluvial deposit that spans
In light of the aforementioned, the purpose of this the entire island of Singapore [9, 33].
study is to develop a slope susceptibility map for Singa- In the Singapore context, the majority of slope failures
pore that takes into account geographical and geotech- occur either during or after rainfall events [5]. This can
nical data in addition to rainfall circumstances, with the be attributed to the infiltration of rainfall, which leads to
ultimate goal of preventing rainfall-induced slope failure an increase in the water content of the soil due to down-
through better assessment of slopes. In addition, this ward flux. As the soil water content increases, the matric
research explores the use of the GeoBarrier System as a suction decreases, resulting in a reduction of soil shear
preventative measure on Singapore’s high-risk slopes. strength [15, 27, 41]. Consequently, residual soil slopes
Slope failures were frequent in the Jurong Formation and that have undergone infiltration are susceptible to failures
Bukit Timah Granite, according to research by Rahardjo triggered by rainfall. In 1995, Rahardjo et al. conducted
et al. [47]. Therefore, the purpose of this research is to a case study to determine the threshold of precipitation
develop the factor of safety distributions for the Jurong that can induce slope failures [46]. The critical limit was
Formation and Bukit Timah Granite in Singapore that found to be a combination of 5 days of antecedent rain-
take into account unsaturated soil properties within fine- fall along with the subsequent precipitation that triggers
scale zonations. The scope of work encompasses regional slope instability.
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 4 of 21

Fig. 1 Distribution of four major formations in Singapore’s geological landscape [52]

In order to conduct infiltration analyses, two rainfall neurons arranged in layers [25]. In an ANN, each node
scenarios can be utilized. The first scenario involves a performs computations using weighted inputs, applies an
1-day maximum daily rainfall of 14.7 mm/hr sustained activation function, and produces an output.
over a 24-h period. The second scenario includes a 5-day To optimize the ANN’s performance, a gradient descent
antecedent rainfall of 2.3 mm/hr for 96 h, followed by a algorithm is commonly employed. The backpropagation
1-day rainfall event of 14.7 mm/hr. The selection of the algorithm, a form of gradient descent, is used to update
appropriate scenario relies on the permeability charac- the weights in the network based on the calculated gradi-
teristics of the soil under consideration. Kim et al. found ent of the error function with respect to the weights [54].
that the Jurong Formation (JF) and Kallang Formation This process iteratively adjusts the weights to minimize
(KF) consist of soil with low permeability, whereas the the difference between predicted and actual soil prop-
Bukit Timah Granite (BTG) and Old Alluvium (OA) con- erty values. The feedforward process in an ANN can be
tain soil with high permeability [26]. To assess slope sta- expressed mathematically as follows:
bility, soils with high permeability are analysed using the
1-day maximum daily rainfall scenario [26]. Conversely, zj = wij ∗ xi + bj (1)
for soils with low permeability, slope stability is evaluated
based on the 5-day antecedent rainfall scenario [26]. where zj is the weighted sum of inputs xi multiplied by
their corresponding weights wij , and bj is the bias term
for the jth node.
3 Applicable theory The activation function introduces non-linearity and
3.1 Artificial Neural Networks (ANN) enables the network to approximate non-linear relation-
Artificial Neural Networks (ANN) have gained sig- ships in the data. One commonly used activation function
nificant popularity in recent years due to their ability to is the Rectified Linear Unit (ReLU), as shown in Eq. (2)
capture complex non-linear relationships between input [40]. The ReLU activation function introduces non-line-
variables and output predictions [20]. The Artificial Neu- arity, enabling the ANN to model and approximate non-
ral Network (ANN) is a computational model that draws linear patterns in the soil property data effectively.
inspiration from the organization and functioning of the
human brain. It is composed of interconnected nodes or f (x) = max (0, x) (2)
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 5 of 21

The activation function f(x) is then applied to the weighted

 � � �2 �
3h 1 h
c0 + c for 0 < h ≤ r

sum zj to obtain the output of the jth node aj : r(h) = 2r − 2 r (7)
c0 + c for h > r
  
aj = f zj (3)
where γ xi − xj is the semivariance between values at
 
The weights and biases of the nodes are adjusted during
location xi and xj , γ (xi − x0 ) is the semivariance between
the training process to minimize the difference between
values at location xj and unmeasured location x0 , and
predicted outputs and actual target values. During the
ψ(x0 ) is the Lagrange multiplier to minimize the error
training phase of the ANN, a learning rate (η) is used to
variance, γ (h) is the estimated semi-variance value for all
control the step size of weight updates. The learning rate
pairs at a lag distance of h, m is the number of pairs sepa-
determines the step size of the weight and bias updates
rated by a distance of h, and Z(xi +h) is the measured value
[31]. A suitable learning rate is crucial for achieving
at the location that is at a distance of h from position xi ,
accurate and stable predictions. A higher learning rate
c0 is the nugget constant, c0 + c is the sill, indicating the
can result in faster convergence, but it also carries the
maximum variability of the variable, and r denotes the
risk of overshooting the optimal solution. Conversely, a
range, which defines the distance beyond which the spa-
lower learning rate may slow down the training process.
tial autocorrelation is negligible.
It is important to choose an appropriate learning rate for
Once the variogram model is fitted to the observed
effective training. The accuracy of an ANN model is com-
semivariance values, it can be used to estimate the semi-
monly evaluated using metrics such as mean absolute
variance for any distance or direction. The semivariance
error (MAE) and root mean squared error (RMSE) for
is then used to calculate the kriging weights, which reflect
regression tasks. These metrics quantify the difference
the spatial relationship between the estimation location
between predicted values and actual values, providing a
and the observed data points. The estimated value at an
measure of the model’s performance.
unmeasured location is obtained by taking a weighted
average of the observed values, with the weights being
3.2 Spatial interpolation using ordinary kriging method
determined by the kriging weights.
The Ordinary Kriging (OK) method is a geostatisti-
cal interpolation technique used to estimate values at
3.3 Regional analyses
unmeasured locations based on the spatial autocorre-
The TRIGRS software provides the capability to simulate
lation of the observed data. This method is used in this
the one-dimensional infiltration of rainwater into the soil,
study since it has been widely employed in various fields
as well as the subsurface flow and runoff resulting from
for its simplicity and availability in many Geographic
precipitation events. It models the soil as a two-layer sys-
Information Systems (GISs) [23]. In OK, the estimation of
tem, comprising an unsaturated zone at the surface and
unknown values is performed by calculating a weighted
a saturated zone with a capillary fringe above the water
average of neighbouring observed values as follows:
table. Ground surface infiltration and vertical flow in the
∗
n unsaturated zone are described by a 1-dimensional ver-
Z(x = i Z(xi ) (4)
0) i=1 sion of the Richards equation, as shown in Eq. (8) [13].
The model incorporates the equations proposed by Gard-
where Z ∗ (x0 ) is the estimated value at the unmeasured
ner in 1958 for the best-fit representation, as illustrated
location x0, Z(xi ) is the measured value at the location xi ,
in Eqs. (9) and (10), respectively [14].
i is the kriging weight for each observed value Z(xi ) and
the sum of the kriging weights is 1 (i.e., ni=1 i = 1), and ∂θ ∂
 
1 ∂ψ

n is the number of positions within the searching neigh- = k(ψ) −1 (8)
∂t ∂z cos2 δ ∂z
bourhood [23].
The determination of kriging weights involves mini-
k(ψ) = ks exp αψ ∗
 
mizing the prediction error variance by incorporating the (9)
spatial dependence information through the variogram
model as shown in Eqs. (5, 6 and 7) [23]: θ = θr + (θs − θr ) exp αψ ∗
 
(10)
n    
i=1
i γ xi − xj + ψ(x0 ) = γ xj − x0 (5) where k(ψ) is the hydraulic conductivity function, δ is the
slope angle, kS is the saturated hydraulic conductivity, α is
a fitting parameter, ψ is the pressure head, ψ ∗ = ψ − ψ0,
1 m  2
r(h) = Z(xi +h) − Z(xi ) (6) where ψ0 is a constant defined as 0 at the water table, θ is
2m i=1
the volumetric water content, θr is the residual water con-
tent, and θs is the saturated water content.
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 6 of 21

After substituting Eqs. 9 and 10 into the one-dimen- for two-dimensional slope stability analyses. SLOPE/W
sional Richards equation, Eqs. 11 and 12 are obtained utilizes Bishop’s simplified technique of slices for assess-
which provide solutions to the resulting linear partial dif- ing slope stability. Bishop’s simplified method assumes
ferential equation. These equations describe the flow at horizontal resultant interslice forces, disregarding inter-
the base of the unsaturated zone at any given time, t. slice shear forces. As a result, Bishop’s simplified tech-
� � � � �  nique ensures moment equilibrium and vertical force
D t
 �IZ − 4(IZ − IZLT ) exp α12du exp − 4ψ equilibrium but does not account for horizontal force
 � 

q(du , t) = ��∞
equilibrium.
�
�m sin(�m α1 du ) 2
� m=1 1+ α1 du +2�2 α1 du exp(−�m Dψ t)
� 
Pressure Plate, Centrifuge, and Tempe cell testing are
 
2 m

(11) used to determine the soil–water characteristic curve

where, [68]. These laboratory studies will provide a few discrete
SWCC data points. A continuous SWCC is required
α1 k s
Dψ = (12) for numerical analysis, hence Fredlund and Xing best-
θs − θr fitting equation is used to generate a continuous curve
where Dψ is the soil–water diffusivity [13], du is the verti- [12]. Zhai and Rahardjo provide a mathematical model
cal depth to the top of the capillary fringe, IZ is surface and methodologies for SWCC variable determination
flux, IZLT is the initial surface flux. Dψ t and α1 du repre- [64]. Kunze et al.’s statistical model estimates the perme-
sent non-dimensional time and depth, respectively, as ability function [29]. The best-fitted SWCC and meas-
used by Srivastava and Yeh [61]. The values of m are ured saturated permeability are needed to estimate the
the positive roots of the pseudoperiodic characteristic unsaturated permeability function. The statistical model’s
of Eq. 13. α1 = αcos2 δ which results from the coordinate accuracy depends on the SWCC’s discretization over its
transformation of the sloping ground surface. TRIGRS entire matric suction range and the number and location
determines values of m using a combination of bisec- of discrete points used to represent the unsaturated soil’s
tion, bracketing, and Newton–Raphson iteration [44]. permeability function [65]. The shear strength of soil is
required for slope stability analyses.
tan(�m α1 du ) + 2� = 0 (13) Shear strength in unsaturated soil was shown to be
significantly impacted only by capillary water in a stress
In addition to TRIGRS, SEEP/W is utilized to con-
analysis conducted by Zhai et al. [66]. Soil suction above
duct 2-D seepage analyses. SEEP/W utilizes a water-
3100 kPa was found to be dominated by hydroscopic
flow governing partial differential equation, as presented
water, or water that is attached to each individual soil
by Fredlund and Rahardjo [11], to solve transient or
particle. In 2020, Zhai et al. developed the following
unsteady-state seepage analyses. This equation serves as
equation to estimate the shear strength of unsaturated
the basis for analysing two-dimensional seepage behav-
soil by removing hydroscopic water in the conventional
iour. Scoops3D is used to execute 3-D slope stability
SWCC [67].
analyses in Singapore to determine the safety factor of
the residual soil slopes. Scoops3D employs the three- S − S′
dimensional limit equilibrium method to assess the safety τ = c′ + (σ − ua )tanφ ′ + (ua − uw )tanφ ′
1 − S′
factor of residual soil slopes. The potential failure mass (14)
consists of three-dimensional vertical columns. Columns where, S is the degree of saturation, S′ is the degree of sat-
along the perimeter of the potential failure mass may be uration corresponding to the suction of 3100 kPa, S is the
partially contained within the sphere. The use of partial degree of saturation corresponding to the suction of ψi,
columns approximates the volume, weight, and slide ua refers to the pore-air pressure and uw is the pore-water
surface area more precisely than using only complete pressure.
columns. To account for the effect of unsaturated soil
conditions, Vanapalli and Fredlund proposed a simpli-
fied approximation of Bishop’s effective stress parameter 3.4 GeoBarrier System (GBS)
[3, 62]. This approximation modifies the pore-water pres- A retaining structure must be taken into considera-
sures in unsaturated soil and is applicable to a general- tion when constructing a slope that is steeper than 45
ized stress law, as described by Lu and Likos [37]. The degrees. Subsequently, the GeoBarrier System (GBS)
simplified approximation offers a representation of the was proposed as a retaining structure that integrates the
soil–water characteristic curve (SWCC) and allows for Capillary Barrier System (CBS) and vegetation [51]. The
its application within a generalized stress law framework. formation of the fine- and coarse-grained layers using
Furthermore, engineers commonly employ SLOPE/W recycled materials instead of natural materials like sand
or gravel is facilitated by the GeoBarrier System (GBS),
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 7 of 21

which subsequently fosters recycling as part of sustain- obtained from the borehole logs. A total of 65 complete
able development. Furthermore, since concrete and steel sets of soil data were used in the ANN training of Jurong
are not required for the GeoBarrier System (GBS), it is Formation with 4 hidden layers and 1000 nodes in each
more environmentally safe and profitable for implemen- layer. The number of iterations used in the training was
tation in both rural and urban areas. Figure 2 illustrates 1000, and the learning rate in the stochastic gradient
that the GeoBarrier System (GBS) is made up of three descent model was 0.1. The error associated with this
main components: reinforced soil system (RSS), CBS, prediction was 9.4%. A total of 185 complete sets of soil
and approved soil mixture (ASM). data were used in the ANN training of Bukit Timah Gran-
Several criteria were considered to evaluate the capil- ite with 4 hidden layers and 1000 nodes in each layer. The
lary barrier’s design. Rahardjo et al. [47] proposed the number of iterations used in the training was 1000, and
following criteria: 1) the water-entry value, ψw or WEV, the learning rate in the stochastic gradient descent model
of the coarse-grained soil should preferably be less than was 0.1. The error associated with this prediction was 5%.
1 kPa; 2) the ratio between the water-entry values of fine- The impact of the prediction error on the soil parameters
grained and coarse-grained materials (ψw-ratio) to be used in the analyses is minimal. Parametric studies have
greater than 10; 3) the saturated permeability coefficient been carried out to investigate this effect and the results
of the non-cohesive fine-grained material (preferably of factor of safety between different prediction results is
greater than ­10–5 m/s). minimal. Hence, the effective cohesion in Jurong Forma-
tion and Bukit Timah Granite was successfully estimated
4 Research methodology using the ANN model based on the index properties (i.e.,
The database for all soil data was set up following pro- grain size distribution and Atterberg limits). After the
cedures suggested by Satyanaga et al. in 2022 [59]. The ANN training was completed, Ordinary Kriging was used
database was developed using the borehole data from to spatially distribute the effective cohesion. The frame-
the Integrated Land Information Service portal which work for the development of the slope susceptibility map
consists of bore logs information from different public in this study is presented in Fig. 3.
agencies in Singapore. A total of 23936 boreholes records For the purpose of performing the regional seepage
containing information such as in-situ water content, (TRIGRS) and slope stability (Scoops3D) analyses, the
specific weight, Atterberg limits, grain size distribution, Jurong formation and Bukit Timah Granite area had to be
saturated permeability, effective cohesion (c’), and effec- split into smaller zones. The raster grid data of the c’ vari-
tive internal friction angle (φ′) were obtained. Rainfall- able over Jurong Formation and Bukit Timah Formation
induced slope failures in Singapore are mostly of shallow were used as the main criteria for delineation of zones as
depth [35], thus the soil properties between 0 to 6 m it is one of the most important soil properties in assess-
depths were used for the database development. Firstly, ing slope stability. In the first step, the classification of c’
ANN training was conducted using the liquid limit, plas- was completed using equal classes of 0.5 kPa. Based on
tic limit, fine grain content, and coarse grain content the spatial distribution of c’, the polygons representing the

Fig. 2 Schematic diagram of GBS

Rahardjo et al. Urban Lifeline (2023) 1:5 Page 8 of 21

Fig. 3 Framework used in this study for the development of slope susceptibility map

zones were completed. The delineation also follows the analyses were prepared manually based on a vector file.
areas with low slope angle, mainly less than 10° in most As a guideline for zones’ delineation, the raster grid of c’
of the cases. These criteria were taken into consideration and slope angle with classes were overlapped. Figure 4
since no instability problems will be encountered within shows the methodology framework applied for the zona-
the areas with a low slope angle. The polygon features tion of Jurong formation and Bukit Timah Granite for the
of the zones in Singapore for TRIGRS and Scoops3D TRIGRS and Scoops3D analyses.

Fig. 4 Schematic diagram with the methodological steps for zonations in Singapore
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 9 of 21

The mean value of all the borehole data inside each stability analyses. Transient seepage analysis was per-
zone was used to calculate the other soil attributes. The formed using TRIGRS software which made use of high-
overall coherence was calculated as the mean of the resolution data on the topography of the small zone. The
zones’ Ordinary Kriging scores. Each zone’s average pore-water pressure distributions from the TRIGRS seep-
borehole would be used to determine the effective fric- age analysis were used as the pore-water pressure condi-
tion angle (φ′), and unit weight. Each zone’s soil type was tions for the slope stability analysis performed using the
reflected by the USCS soil classification that received Scoops3D software. The respective input maps such as
the most votes from boreholes located within that zone. digital elevation model (DEM), slope map, groundwater
Typical hydraulic parameters were assigned to each table (GWT) map and flow map as shown in Figs. 5 and
soil type after determining the representative soil type 6 were used to define the topography and initial ground-
in each zone, as recommended by Mercer et al. (2019). water table conditions in the TRIGRS and Scoops3D
The hydraulic properties of soil, such as the SWCC and analyses. The maximum daily and maximum 5-day cumu-
unsaturated permeability, were estimated based on the lative rainfall from historical periods between 1982 and
soil type since the borehole records do not contain all the 2017 were used to calculate the extreme rainfall loading
essential soil properties. Since TRIGRS uses the Gard- of 577 mm/5 day and 353 mm/day which were then
ner’s equation to calculate the SWCC, the SWCC from applied in the analyses [51]. Rahardjo et al. and Rahimi
the Fredlund & Xing equation was best fitted based on et al. found that residual soil slopes with low permeability
the minimum root-mean-square error to determine the soil attained a lower factor of safety when subjected to
Gardner’s fitting parameter (α) [12]. an antecedent rainfall [48, 53]. Therefore, the JF17 zone
One zone from Jurong Formation (i.e. JF17) and Bukit which has a low saturated permeability of 5E-07 m/s was
Timah Granite (i.e. BTG29) which had experienced subjected to a constant 5-day antecedent rainfall loading of
past slope failure was selected for the seepage and slope 115 mm/day for a duration of 5 days followed by another

Fig. 5 JF17 input maps (a. DEM; b. Slope; c. Groundwater table; d. flow) used in TRIGRS and Scoops3D
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 10 of 21

Fig. 6 BTG29 input maps (a. DEM; b. Slope; c. Groundwater table; d. flow) used in TRIGRS and Scoops3D

5 days of no rainfall to analyse the recovery of the factor of the past and thus were selected for the 2D analysis. The
safety. While BTG29 which has a high saturated perme- Bukit Teresa and Emerald Hill slopes were selected as real-
ability of 1.00E-06 m/s was subjected to 24 h of maximum time monitoring study was conducted by instrumenting
daily rainfall intensity of 353mm/day followed by another these slopes with moisture sensors. Hence, the actual soil
24 h of no rainfall. properties of these slopes were available and used for the
The computed factor of safety (FoS) distribution from 2D analyses. The location of the six slopes are depicted in
the Scoops3D analyses was used to create a suscep- Figs. 5 and 6. The soil properties used in the TRIGRS and
tibility map for slopes. High risk (FoS ≤ 1), moderate Scoops3D analyses were also applied for the 2D analyses
risk (FoS ≤ 1.25), low risk (FoS ≤ 1.5), and very low risk for the slopes with past failure history. However, the soil
(FoS > 1.5) were the four categories used to differentiate properties obtained from the laboratory tests were used
the map into four risk areas. The slope susceptibility map for the instrumented slopes (i.e. Bukit Teresa and Emer-
was then used to pinpoint high-risk locations within the ald Hill). Unsaturated and saturated laboratory tests were
zone, prompting the follow-up with 2-dimensional seep-
age and stability assessments.
Two-dimensional (2D) seepage and slope stability anal- Table 1 Slope properties used in the numerical analyses
yses using the GeoStudio 2012 software, SEEP/W and
SLOPE/W, were performed for three residual soil slopes Slope location Zone Slope Slope Water Water
Height Angle Table at Table at
from the JF17 zone (i.e. Bukit Teresa, Jalan Minyak, Lower (m) (o) Crest (m) Toe (m)
Delta) and three residual soil slopes from BTG29 zones
(i.e. Emerald Hill, Wilkie Road, Waterloo Street). The pur- Bukit Teresa JF17 5 27 11 6
pose of the 2D analyses is to compare and verify the results Emerald Hill BTG29 3 30 3 1
from the 3D TRIGRS and Scoops3D analyses. Hence, the Jalan Minyak JF17 4 45 5 1
same rainfall intensity and duration were applied in the 2D Lower Delta JF17 8 40 9 6
analyses. The slopes at Jalan Minyak, Lower Delta, Wilkie Wilkie Road BTG29 5 35 7 3
Road, and Waterloo Street experienced slope failures in Waterloo Street BTG29 3 30 3 1
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 11 of 21

Fig. 7 Typical slope geometry and boundary conditions applied in the 2D analysis

conducted on undisturbed soil samples from 3–4 m depth height of the slope from the crest and the toe to exclude
to obtain the relevant soil properties such as SWCC, per- the effect of the boundary conditions on the seep-
meability function and shear strength properties required age within the slope. A constant total head boundary
for the seepage and slope stability analyses. condition was applied on the right and left boundaries
Table 1 shows the slope height, slope angle and while a no flow condition without seepage face review
groundwater table depth at the crest and toe for all six was applied to the bottom boundary. To simulate rain-
slopes. The typical slope geometry and boundary con- fall infiltration without any ponding, a unit flux bound-
ditions modelled in the 2D analysis are presented in ary with potential seepage face review was applied to
Fig. 7. The side boundaries have a length of 3 times the the surface of the slope.

Fig. 8 Zonation of the Residual Soils from Jurong Formation and Bukit Timah Granite Formations
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 12 of 21

Table 2 Soil properties from soil database

Geological Formation Zone USCS Permeability Unsaturated Properties Effective cohesion Effective
friction
angle
ks (m/s) Diffusivity Alpha c ′(kPa) φ ′(o)

Jurong Formation JF1 ML 5E-06 4.9E-05 0.012 9.5 31

JF2 CL 1E-06 4.9E-06 0.00086 9.5 31
JF3 CL 1E-06 4.9E-06 0.00086 9 31
JF4 CL 1E-06 4.9E-06 0.00086 9.5 31
JF5 CL 1E-06 4.9E-06 0.00086 9 31
JF6 CL 1E-06 4.9E-06 0.00086 9.5 31
JF7 MH 1E-06 2.7E-06 0.00124 10 31
JF8 CL 1E-06 4.9E-06 0.00086 10 31
JF9 CL 1E-06 4.9E-06 0.00086 9.5 31
JF10 CL 1E-06 4.9E-06 0.00086 9.5 31
JF11 CL 1E-06 4.9E-06 0.00086 9 31
JF12 CL 1E-06 4.9E-06 0.00086 12 29
JF13 CL 1E-06 4.9E-06 0.00086 9 31
JF14 CL 1E-06 2.7E-06 0.00124 9.5 31
JF15 CL 1E-06 4.9E-06 0.00086 9 31
JF16 CH 5E-07 2.4E-06 0.00124 9 31
JF17 CH 5E-07 2.4E-06 0.00124 10 31
Bukit Timah Granite BTG1 MH 5.00E-06 1.65E-05 0.01588 8.65 31
BTG2 CL 1.00E-06 4.91E-06 0.00864 9.77 34
BTG3 MH 5.00E-06 1.65E-05 0.01588 9.16 34
BTG4 SM 5.00E-05 2.00E-04 0.02013 7.88 34
BTG5 SM 5.00E-05 2.00E-04 0.02013 8.42 33
BTG6 MH 5.00E-06 1.65E-05 0.01588 10.18 36
BTG7 MH 5.00E-06 1.65E-05 0.01588 8 33
BTG8 MH 5.00E-06 1.65E-05 0.01588 9.44 34
BTG9 MH 5.00E-06 1.65E-05 0.01588 8.52 33
BTG10 MH 5.00E-06 1.65E-05 0.01588 7.85 36
BTG11 MH 5.00E-06 1.65E-05 0.01588 9.83 34
BTG12 MH 5.00E-06 1.65E-05 0.01588 7.28 32
BTG13 MH 5.00E-06 1.65E-05 0.01588 7.28 32
BTG14 GC 1.00E-04 1.29E-04 0.29543 8.52 32
BTG15 SC 1.00E-05 6.34E-05 0.08580 9.09 34
BTG16 CL 1.00E-06 4.91E-06 0.00864 7.38 32
BTG17 MH 5.00E-06 1.65E-05 0.01588 8.34 30
BTG18 SC 1.00E-05 6.34E-05 0.08580 6.59 33
BTG19 SM 5.00E-05 2.00E-04 0.02013 8.43 34
BTG20 MH 5.00E-06 1.65E-05 0.01588 8.01 31
BTG21 MH 5.00E-06 1.65E-05 0.01588 7.88 35
BTG22 CH 1.00E-06 4.37E-06 0.006736 9.35 32
BTG23 CH 1.00E-06 4.37E-06 0.00674 9.33 30
BTG24 CH 1.00E-06 4.37E-06 0.00674 9.41 29
BTG25 MH 5.00E-06 1.65E-05 0.01588 9.61 32
BTG26 CH 1.00E-06 4.37E-06 0.00674 8.02 25
BTG27 MH 5.00E-06 1.65E-05 0.01588 9.52 30
BTG28 MH 5.00E-06 1.65E-05 0.01588 9.79 27
BTG29 CH 1.00E-06 4.37E-06 0.00674 9.27 28
BTG30 MH 5.00E-06 1.65E-05 0.01588 9.04 28
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 13 of 21

5 Slope susceptibility map obtained from the Scoops3D analyses are presented in
After the c’ spatial variation in Jurong Formation and Bukit Fig. 9. Five out of six investigated slopes obtained a fac-
Timah Granite area had been developed, the formation tor of safety between 1.25 and 1.5 indicating these slopes
was divided into smaller zones. The Jurong Formation was are of moderate risk.
divided into 16 zones with each zone having a different soil The results from the laboratory tests performed on the
property. Thereafter, bigger zones were further divided, soil samples obtained from Bukit Teresa and Emerald Hill
forming 32 polygons for the numerical analyses. Mean- slopes are presented in Fig. 10. The laboratory data points
while, the Bukit Timah Granite area was divided into 28 from the SWCC tests were best fitted using the Fred-
zones and thereafter into 39 polygons. The polygons and lund and Xing equation to obtain a continuous SWCC as
zones for both formations are presented in Fig. 8. shown in Fig. 10 [12]. The unsaturated permeability func-
The mean c’ for each zone obtained from the ANN tions presented in Fig. 11 were calculated by the statisti-
analysis are presented in Table 2. The other soil prop- cal method proposed by Kunze et al. using the saturated
erties such as saturated permeability, unsaturated soil coefficient of permeability and the SWCC and thereafter
properties such as Gardner SWCC parameter alpha and best fitted using the equation proposed by Leong and
diffusivity and shear strength properties are also pre- Rahardjo in 1997 [29, 32].
sented in Table 2. Satyanaga and Rahardjo conducted The factor of safety variation with time from the 2D
analyses to develop a slope susceptibility map with and analyses for the six investigated slopes are presented in
without the incorporation of unsaturated soil properties Fig. 12. In general, slopes in both formations attained
[57]. They observed that the utilisation of the saturated the lowest FoS at the end of the rainfall and thereafter
soil properties in the development of the slope suscepti- an increase in FoS was observed during the no rain-
bility map resulted in extensive areas with slopes having fall period. However, the factor of safety at the end of
a FoS of less than 1.5. This may result in unreasonably the no rainfall period did not recover to the initial FoS
high costs being required for slope preventative meas- before the start of the rainfall. The FoS maps obtained
ures and/or slope repairs. However, upon close obser- from TRIGRS and Scoops3D analyses were verified
vation, slopes located within the area with a FoS of less with the 2D numerical analyses from SEEP/W and
than 1.5 appear to be stable without any slope improve- SLOPE/W. The comparison of the FoS between the two
ment. In other words, those slopes are actually unsatu- approaches for the six investigated slopes are presented
rated soil slopes where matric suctions contribute to the in Table 3.
shear strength and stability of the slopes. As a result, the The FoS obtained from TRIGRS (1-D infiltration) and
FoS values of these slopes are in reality higher than 1.5 Scoops3D (3-D slope stability) were found to be lower
and can be considered safe. These slopes will not require (more conservative) compared to the results obtained
any slope improvement work. In summary, the gener- using SEEP/W and SLOPE/W (2-D seepage and slope
ated slope susceptibility map with the assumption of a stability). This could be due to the fact that the 1-D infil-
fully saturated slope was unsustainable for infrastructure tration of the TRIGRS model produced higher pore-
development. Based on this study, the unsaturated soil water pressures as compared to the 2-D seepage analysis
properties are incorporated in the TRIGRS analyses. The from SEEP/W. As a result, the higher pore-water pres-
factor of safety (FoS) maps for JF17 and BTG29 zones sures contribute to a lower FoS.

Fig. 9 Factor of safety map of (a) JF17; (b) BTG29 zone

Rahardjo et al. Urban Lifeline (2023) 1:5 Page 14 of 21

Fig. 10 SWCC for Bukit Teresa and Emerald Hill soils

6 Geobarrier system the capillary barrier effect evaluation criteria using the
The soil–water characteristics curves (SWCCs) of RCA parameters obtained from Figs. 13 and 14. As shown in
and ASM are presented in Fig. 13. Figure 14 shows the Fig. 13, the water-entry values of fine RCA and coarse
permeability functions of RCA and ASM determined RCA are 4 kPa and 0.17 kPa, respectively. The water
from SWCC and saturated permeability using the statis- entry value of coarse-grained RCA (0.17 kPa) satisfied the
tical method following the procedure explained by Saty- requirement for coarse-grained material in CBS (< 1 kPa).
anaga et al. in 2022 [59]. Table 4 presents a summary of The ratio of the water-entry value of fine RCA to coarse

Fig. 11 Permeability function for Bukit Teresa and Emerald Hill soils
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 15 of 21

Fig. 12 FoS results from 2D analyses for the 6 investigated slopes (a) JF17 zone; (b) BTG29 zone

Table 3 Comparison of FoS obtained from map and 2-D Figure 15 depicts the boundary conditions of the
numerical model Emerald Hill slope and key components of the GBS. In
the finite element analysis, the element size was 0.1 m.
Location FoS from slope FoS from
susceptibility map SEEP/W and The GBS comprised residual soil, geogrid reinforce-
SLOPE/W ment, fine and coarse RCA for the CBS, and ASM for
the green cover. The thicknesses of each of the GBS com-
Bukit Teresa 1.41 1.62
ponents are as follows: fine RCA – 50 cm, coarse RCA
Emerald Hill 1.47 1.62
– 30 cm, ASM – 60 cm. GBS consists of six lifts (bag
Jalan Minyak 1.35 1.53
height is 50 cm) for a 3 m slope. The requirement for the
Lower Delta 1.41 1.65
minimum length of geogrid is 70% of the slope height.
Wilkie Road 1.29 1.41
Accordingly, the geogrid length was taken as 2.1 m.
Waterloo Street 1.92 2.12
The analyses assigned an impermeable layer (a separa-
tor) behind the GBS as a no-flow boundary condition.
RCA (23.5) met the criterion for the ratio of fine-grained Changes in pore-water pressure were incorporated into
to coarse-grained materials (≥ 10). The measured satu- slope stability analyses as a result of seepage analyses.
rated permeability of fine RCA (2.7 × ­10−4 m/s) met CBS The analysis was performed by selecting the Mohr–Cou-
criterion (> ­10−5 m/s). lomb material model and inputting the shear strength
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 16 of 21

Fig. 13 SWCCs of RCA and ASM

Fig. 14 Permeability functions of RCA and ASM

Table 4 Check of the design criteria for a capillary barrier system parameters for each material, which can be found in
to be effective Table 5.
Parameter Fine RCA​ Coarse RCA​ As seen from Table 6, the minimum factor of safety
of the Emerald Hill slope without GBS is 1.62, which is
Water-entry value, ψw (kPa) 4.0 0.17 (< 1) lower than the factor of safety of a slope with the cover
ψw-ratio 23.5 (> 10) system included, which is 1.74, as shown in Fig. 16.
Saturated permeability coef- 2.7 × ­10–4 (> ­10–5) 7.5 × ­10–3 Figure 17 illustrates the change in factor of safety
ficient, ks (m/s)
against time during and after rainfall for the Emerald Hill
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 17 of 21

Fig. 15 A numerical model for seepage analyses: GBS covering Emerald Hill slope with a 3 m height and a 70 (°) slope angle

Table 5 Summary of shear strength parameters of the materials involved inside the GBS
Symbol Unit Residual soil Coarse RCA​ Fine RCA​ ASM

Effective Cohesion c’ kPa 3 0 0 2

Effective Internal Friction Angle φ’ degree (°) 35 35 34 30
Unsaturated Shear Strength Angle φb degree (°) 17.5 17.5 17 15
Unit Weight γ kN/m3 18 21 20 18

slope with and without the GBS. The decrease in the fac- the Emerald Hill slope, the mesh size in the finite ele-
tor of safety when the slope was covered with GBS was ment analysis for the Bukit Teresa slope was also 0.1 m.
lower as compared to the slope without cover. This may The GBS was made up of residual soil, geogrid rein-
be attributed to the effectiveness of the GBS in minimal- forcement, coarse and fine RCA for the CBS, and ASM
izing the rainfall infiltration into the slope. It should be for the green cover. The same SWCCs and permeability
noted that the initial factor of safety of the Emerald Hill functions of the materials shown in Figs. 10 and 11 were
slope without the cover was 2.326 and decreased to a assigned in this analysis. Each of the GBS components
value of 1.623 (i.e., ∆FoS = 0.7 or 30% reduction). While, has the following thicknesses: coarse RCA – 30 cm, fine
the initial factor of safety of the slope with the GBS RCA – 50 cm, ASM – 60 cm. GBS consists of ten lifts
cover was 1.905 and decreased to a value of 1.739 (i.e., (bag height is 50 cm) for a 5 m slope. The requirement for
∆FoS = 0.17 or 8% reduction). the minimum length of geogrid is 70% of the slope height.
Figure 18 depicts the boundary conditions of the Bukit Accordingly, the geogrid length was taken as 3.5 m. This
Teresa slope and the main elements of the GBS. As with study is limited to investigate the effect of GBS in pre-
venting rainwater infiltration from the slope surface.
It is assumed there is no water infiltration from the soil
Table 6 Factor of safety of the Emerald Hill slope with and behind the compacted soil of GBS. In addition, the actual
without GBS at different time periods implementation of GBS in Singapore incorporates the
Without GBS With GBS
impermeable layer behind the compacted soil of GBS to
prevent the water leaking from the water pipe within the
Initial FOS 2.326 1.905 soil behind GBS to the GBS area. Therefore, the analyses
FOS at the end of the rainfall 1.623 (min) 1.762 designated an impermeable layer as a no-flow boundary
Final FOS (at the end 1.717 1.739 (min) condition behind the GBS. As a result of seepage stud-
of entire period)
ies, changes in pore-water pressure were integrated into
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 18 of 21

Fig. 16 Minimum factor of safety and a critical slip surface of Emerald Hill slope covered with the GBS

Fig. 17 Changes in the factor of safety versus time of the Emerald Hill slope with and without GBS during and after 1-day maximum rainfall

Fig. 18 A numerical model for seepage analyses: GBS covering Bukit Teresa slope with a 5 m height and a 70(°) slope angle
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 19 of 21

the slope stability analyses. The study was carried out by be made steeper, thereby saving urban space. In addi-
selecting the Mohr–Coulomb material model and input- tion, a capillary barrier system can prevent rainfall infil-
ting the shear strength parameters for each material tration, and the steep slope remains stable.
shown in Table 5.
The minimum factor of safety of the Bukit Teresa slope 7 Conclusions
with GBS, including its corresponding critical slip sur- The following conclusions can be drawn from this study:
face is presented in Fig. 19. The minimum factor safety is
equal to 2.31. Table 7 shows the key values for the factors 1. Slope susceptibility map was generated using TRI-
of safety of the Bukit Teresa slope with and without GBS. GRS and Scoops3D using soil properties obtained
The minimum factor of safety of the Bukit Teresa slope from Rahardjo et al. and using extreme rainfall case
without GBS is 2.84, and it is higher than the slope’s (577 mm/5 days for Jurong Formation and 353 mm/
safety factor when the cover system is included, which day for Bukit Timah Granite) [51]. Slope susceptibil-
is 2.31 as shown in Fig. 20. Moreover, the initial factor ity maps allow quick identification of locations that
of safety of the original slope is 4.74, while the slope are susceptible to rainfall-induced slope failures and
with GBS has the initial factor of safety of 3.71. A pos- take remedial actions, especially if these areas are
sible reason for that is that the slope without GBS has near urban infrastructures.
a lower angle than the slope with GBS. Despite this, the 2. Emerald Hill residual soil slope at Bukit Timah Gran-
difference in the factor of safety of a slope covered with ite Formation has a higher percentage of fines, thus
GBS is lower by almost 25%, thus, it can be said that GBS it has a higher air-entry value and lower saturated
contributed to the reduction of rainfall infiltration. Even permeability compared to Bukit Teresa residual
if the factors of safety of two original slopes are higher soil slope at Jurong Formation. The effective fric-
than 1, the use of GBS is preferable since slopes covered tion angle and the effective stress of Bukit Teresa are
with GBS, most importantly, promote sustainability and higher compared to Emerald Hill due to the higher
manage urban space shortages, i.e., slopes with GBS can percentage of clay in Emerald Hill.

Fig. 19 Minimum factor of safety and a critical slip surface of Bukit Teresa slope covered with the GBS

3. Under the extreme rainfall case, the factor of safety

Table 7 Factor of safety of the Bukit Teresa slope with and for Bukit Teresa slope is 1.620 and the factor of safety
without GBS at different time periods for Emerald Hill slope is 1.616.
4. The slope susceptibility map agreed closely with the
Without GBS With GBS
results obtained from SEEP/W and SLOPE/W. Slope
Initial FOS 4.743 3.709 susceptibility map tends to have a slightly lower fac-
FOS at the end of the rainfall 2.839 (min) 2.310 (min) tor of safety which is more conservative.
Final FOS (at the end 3.052 2.388 5. The slope protected with GBS showed a smaller
of entire period) change in FoS during rainfall as compared to the
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 20 of 21

Fig. 20 Changes in the factor of safety versus time of Bukit Teresa slope with and without GBS during and after 5-days maximum rainfall

original slope without GBS. It can be concluded that References

1. Bai S, Lü G, Wang J, Zhou P, Ding L (2011) GIS-based rare events logistic
the application of GBS will provide a successful pro- regression or landslide-susceptibility mapping of Lianyungang, China.
tection method against rainfall-induced landslides. Environ Earth Sci 62:139–149
The application of GBS is not only limited to tropical 2. Baum RL, Savage WZ, Godt JW (2008) TRIGRS—a Fortran program for
transient rainfall infiltration and grid-based regional slope-stability analy-
countries, but it can be used in different regions in sis. US Geological Survey Open-File Rep 424:08–1159
the world with additional research works for imple- 3. Bishop AW (1959) The principle of effective stress. Teknisk Ukeblad 39:859–863
mentation in the cold weather. 4. Chan YE, Ng QL, Satyanaga A, Rahardjo H (2021) Regional stability and
adaptation measures slope failures due to rainfall in Singapore. Environ
Geotech. https://​doi.​org/​10.​1680/​jenge.​20.​00089
5. Chatterjea K (1989) Observations on the fluvial and slope processes in
Acknowledgements
Singapore and their impact on the urban environment. Dissertation,
The authors gratefully acknowledge the assistance of Mr. Andrew Tirtaward-
National University of Singapore
hana, Assel Bayasheva, Aizhan Sagu, Yerkezhan Makenova, Gerarldo Davin in
6. Chigira M (2001) Micro-sheeting of granite and its relationship with
the data collection used in this study.
landsliding specifically after the heavy rainstorm in June 1999, Hiroshima
Prefecture, Japan Eng Geol 59(3–4):219–231
Authors’ contributions
7. Cruden D (1991) A simple definition of a landslide. Bull Eng Geol Env 43(1):27–29
Harianto Rahardjo: Writing - Original Draft, Validation, Formal analysis,
8. Dai F, Lee C (2001) Frequency–volume relation and prediction of rainfall-
Investigation, Data Curation. Zhai Qian: Writing - Original Draft, Conceptualiza-
induced landslides. Eng Geol 59(3):253–266
tion, Methodology, Supervision, Project administration, Funding acquisition.
9. Department PW (1976) Geology of the Republic of Singapore. Public
Alfrendo Satyanaga: Validation, Formal analysis, Investigation, Data Curation.
Works Department, Singapore
Yangyang Li: Conceptualization, Methodology, Data Curation. Saranya Ranga-
10. Farahmand A, Aghakouchak A (2013) A satellite-based global landslide
rajan: Formal analysis, Methodology, Data Curation. Arezoo Rahimi: Review &
model. Nat Hazards Earth Syst Sci 13(5):1259–1267. https://​doi.​org/​10.​
Editing, Supervision, Validation.
5194/​nhess-​13-​1259-​2013
11. Fredlund DG, Rahardjo H (1993) Soil Mechanics for unsaturated soils.
Funding
John Wiley and Sons Inc, New York
The authors received no financial support for the authorship and publication
12. Fredlund DG, Xing A (1994) Equations for the soil–water characteristic
of this article.
curve. Can Geotech J 31(3):521–532
13. Freeze R, Cherry J (1979) Groundwater. Prentice-Hall, Englewood Cliffs, NJ
Availability of data and materials
14. Gardner WR (1958) Some steady-state solutions of the unsaturated
The datasets generated during and/or analysed during the current study are
moisture flow equation with application to evaporation from a water
available from the corresponding author on reasonable request.
table. Soil Sci 85(4):228–232
15. Gitirana G, Fredlund MD (2005) Inflitration–runoff boundary conditions
Declarations in seepage analysis. In: 58th Canadian Geotechnical Conference and
­6th Joint IAH-CGS Conference, Saskatoon, SK, Canada
Competing interests 16. Gu X, Wang L, Ou Q, Zhang W (2023) Efficient stochastic analysis of
The authors declare that they have no known competing financial interests unsaturated slopes subjected to various rainfall intensities and pat-
or personal relationships that could have appeared to influence the work terns. Geosci Front 14(1):101490
reported in this paper. 17. Gu X, Wang L, Ou Q, Zhang W, Sun G (2023) Reliability Assessment of
Rainfall-Induced Slope Stability Using Chebyshev-Galerkin-KL Expan-
sion and Bayesian Approach. Can Geotech J. https://​doi.​org/​10.​1139/​
Received: 14 July 2023 Revised: 4 September 2023 Accepted: 19 Septem- cgj-​2022-​0671
ber 2023 18. Guzzetti F, Peruccacci S, Rossi M, Stark CP (2008) The rainfall intensity–
duration control of shallow landslides and debris flows: an update.
Landslides 5(1):3–17
Rahardjo et al. Urban Lifeline (2023) 1:5 Page 21 of 21

19. Hamdany AH, Wijaya M, Satyanaga A, Rahardjo H, Qian Z, Lim A, Kim J 46. Rahardjo H, Li XW, Toll DG, Leong EC (2001) The effect of antecedent
(2023) Numerical simulation on the effect of infiltration and evapotran- rainfall on slope stability. J Geotech Geolog Eng 19(3–4):371–399
spiration on the residual slope. Sustainability 15(11):8653 47. Rahardjo H, Krisdani H, Leong EC (2007) Application of unsaturated soil
20. Haykin S (1999) Neural networks: A comprehensive foundation, 2nd mechanics in capillary barrier system. In: INVITED LECTURE. Proceed-
edn. Prentice Hall, Hoboken ings of 3rd Asian Conference on Unsaturated Soils., Nanjing, China, pp
21. Hu XL, Tang HM (2005) Research on the GIS system of slope engineer- 127–137
ing GIS and its application. China University of Geosciences Press, 48. Rahardjo H, Leong EC, Rezaur RB (2008) Effect of antecedent rainfall on
Wuhan (in Chinese) pore-water pressure distribution characteristics in residual soil slopes
22. Huabin W, Gangjun L, Weiya X, Gonghui W (2005) GIS-based landslide under tropical rainfall. Hydrol Process 22(4):506–523
hazard assessment: an overview. Prog Phys Geogr 29(4):548–567 49. Rahardjo H, Satyanaga A, Leong EC, Ng YS, Pang HT (2012) Variability of
23. Ip SC, Satyanaga A, Rahardjo H (2021) Spatial variation of shear strength residual soil properties. Eng Geol 141:124–140
properties incorporating auxiliary variables. CATENA 200:105196 50. Rahardjo H, Gofar N, Satyanaga A (2018) Effect of concrete waste particles
24. Kim J, Lee K, Jeong S, Kim G (2014) GIS-based prediction method of land- on infiltration characteristics of soil. Environ Earth Sci 77:1–12
slide susceptibility using a rainfall infiltration-groundwater flow model. 51. Rahardjo H, Satyanaga A, Gofar N, Leong EC, Kew JH, Wang CL, Wong JL
Eng Geol 182:63–78 (2019) Geobarrier system for protection against rainfall-induced slope
25. Kim Y, Satyanaga A, Rahardjo H, Park H, Sham AW (2021) Estimation of failure. ISSMGE Intern J Geoeng Case Histories 5(1):26–42
effective cohesion using artificial neural networks based on index soil 52. Rahardjo H, Leong EC, Qin X, Satyanaga A, Nistor M, Yongmin K, Koh
properties: a Singapore case. Eng Geol 289:106163 ZH (2020) The development of slope management and susceptibility
26. Kim Y, Rahardjo H, Nistor MM, Satyanaga A, Leong EC, Sham AW (2022) geographical information system. Progress Rep 5:60
Assessment of critical rainfall scenarios for slope stability analyses based 53. Rahimi A, Rahardjo H, Leong EC (2011) Effect of antecedent rainfall
on historical rainfall records in Singapore. Environ Earth Sci 81(2):39 patterns on rainfall-induced slope failure. J Geotech Geoenviron Eng
27. Krahn J, Fredlund DG, Klassen MJ (1989) Effect of soil suction on slope 137(5):483–491
stability at Notch Hill. Canadian Geotech J 26(2):269–278 54. Rumelhart D, Hinton G, Williams R (1986) Learning internal representa-
28. Krisnanto S, Rahardjo H, Kartiko RD, Satyanaga A, Nugroho J, Mulyanto tions by backpropagating errors. Nature 323:533–536
N, Santoso P, Hendiarto A, Pamuji DB, Rachma SN (2021) Characteristics 55. Sassa K, Picarelli L, Yin Y (2008) Monitoring, prediction and early warn-
of rainfall-induced slope instability in Cisokan Region Indonesia. J Eng ing. In: Sassa K, Canuti P (eds) Landslides – Disaster Risk Reduction.
Technol 53(5):Sciences (JETS) Springer, Berlin, Heidelberg, pp 351–375. https://​doi.​org/​10.​1007/​978-3-​
29. Kunze RJ, Uehara G, Graham K (1968) Factors important in the calculation 540-​69970-5_​20
of hydraulic conductivity. Soil Sci Soc Am J 32(6):760–765 56. Satyanaga A, Aventian GD, Makenova Y, Zhakiyeva A, Kamaliyeva Z,
30. Larsen MC, Wieczoreck GF, Eaton LS, Morgan BA, Torres-Sierra H (2001) Moon SW, Kim J (2023) Building information modelling for application in
Natural hazards on alluvial fans: the Venezuela debris flow and flash flood geotechnical engineering. Infrastructures 8(6):103
disaster. In: USGS Fact Sheet. FS, vol 103. p 01 57. Satyanaga A, Rahardjo H (2022) Role of unsaturated soil properties in the
31. LeCun Y, Bottou L, Bengio Y, Haffner P (1998) Gradient-based learning development of slope susceptibility map. Proc Inst Civil Eng Geotech
applied to document recognition. Proc IEEE 86(11):2278–2324 Eng 175(3):276–288
32. Leong EC, Rahardjo H (1997) Review of soil-water characteristic curve 58. Satyanaga A, Rahardjo H, Hua CJ (2019) Numerical simulation of capillary
equations. J Geotech Geoenviron Eng 123(12):1106–1117 barrier system under rainfall infiltration in Singapore. ISSMGE Intern J
33. Leong EC, Rahardjo H, Tang SK (2002) Characterization and engineering Geoeng Case Histories 5(1):43–54
properties of Singapore residual soils. In: Proceedings of the International 59. Satyanaga A, Rangarajan S, Rahardjo H, Li Y, Kim Y (2022) Soil database for
Workshop on Characterisation and Engineering Properties of Natural development of soil properties envelope. Eng Geol 304:106698
Soils, Singapore, pp 1–10 60. Sharipov A, Satyanaga A, Abishev R, Moon SW, Taib AM, Kim J (2023) Influ-
34. Lepore C, Arnone E, Noto LV, Sivandran G, Bras RL (2013) Physically based ence of slope geometry on stability of clayey soil slopes. Geotech Geolog
modeling of rainfall-triggered landslides: a case study in the Luquillo for- Eng 41(5):2939–50
est, Puerto Rico. Hydrol Earth Syst Sci 17(9):3371–3387 61. Srivastava R, Yeh TC (1991) Analytical solutions for one-dimensional,
35. Li Y, Rahardjo H, Satyanaga A, Rangarajan S, Lee DT (2022) Soil database transient infiltration toward the water table in homogeneous and layered
development with the application of machine learning methods in soil soils. Water Resour Res 27(5):753–762
properties prediction. Eng Geol 306:106769 62. Vanapalli SK, Fredlund DG (2000) Comparison of different procedures to
36. Liu S, Wang L, Zhang W, Sun W, Fu J, Xiao T, Dai Z (2023) A physics- predict unsaturated soil shear strength. In: Shackelford CD, Houston SL,
informed data-driven model for landslide susceptibility assessment in the Chang N-Y (eds) Advances in Unsaturated Geotechnics. Proceedings of
Three Gorges Reservoir Area. Geosci Fronts 14(5):101621 sessions at Geo-Denver 2000, held in Denver, Colorado, August 5-8, 2000.
37. Lu N, Likos W (2004) Unsaturated soil mechanics. Wiley, New York ASCE, Reston, pp 195–209. https://​doi.​org/​10.​1061/​40510​(287)​13
38. Montgomery DR, Dietrich WE (1994) A physically based model for the top- 63. Yagi H, Maruo Y, Saijo K, Nakamura S (1988) The Sept. 1988 large landslide
ographic control on shallow landsliding. Water Resour Res 30:1153–1171 in the vicinity of MCT, Darbang, Nepal. Landslides 26(4):45–49
39. Montrasio L, Valentino R, Losi GL (2011) Towards a real-time susceptibility 64. Zhai Q, Rahardjo H (2012) Determination of soil–water characteristic
assessment of rainfall-induced shallow landslides on a regional scale. Nat curve variables. Comput Geotech 42:37–43
Hazard 11(7):1927–1947 65. Zhai Q, Rahardjo H (2015) Estimation of permeability function from the
40. Nair V, Hinton GE (2010) Rectified linear units improve restricted boltz- soil–water characteristic curve. Eng Geol 199:148–156
mann machines. In Proceedings of the 27th international conference on 66. Zhai Q, Rahardjo H, Satyanaga A, Dai G (2019) Estimation of unsatu-
machine learning (ICML-10), pp. 807–814 rated shear strength from soil-water characteristic curve. Acta Geotech
41. Ng CW, Shi Q (1998) A numerical investigation of the stability of unsaturated 14:1977–1990
soil slopes subjected to transient seepage. Comput Geotech 22(1):1–28 67. Zhai Q, Rahardjo H, Satyanaga A, Dai G (2020) Estimation of tensile
42. Nistor MM, Rahardjo H, Satyanaga A, Leong EC, Hao KZ, Sham AW, strength of sandy soil from soil-water characteristic curve. Acta Geotech
Hongjun WU (2019) GIS-based approach to identify the suitable locations 15(12):3371–3381
for soil sampling in Singapore. Geogr Tech 14(1):103–117 68. Zhai Q, Xiang K, Rahardjo H, Satyanaga A, Dai G, Gong W, Zhao X (2023) A
43. Pack RT, Tarboton DG, Goodwin CN (1998) The SINMAP approach to ter- new domain model for estimating water distribution in soil pores during
rain stability mapping. In: 8th congress of the international association of the drying and wetting processes. Eng Geol 322:107180
engineering geology, vol 21. Vancouver, British Columbia, Canada, p 25
44. Press WH, Flannery BP, Teukolsky SA, Vetterling WT (1986) Numerical reci-
pes, the art of scientific computing. Cambridge University Press, New York Publisher’s Note
45. Rahardjo H, Satyanaga A (2019) Sensing and monitoring for assessment Springer Nature remains neutral with regard to jurisdictional claims in pub-
of rainfall-induced slope failures in residual soil. Proceed Institution Civil lished maps and institutional affiliations.
Eng Geotech Eng 172(6):496–506