Volume 65 Number 2

June 2023

QQ Spalling of concrete in fire – an overview and local relevance

QQ Design criteria for upstream raised tailings storage facilities
QQ  ould a one-size-fits‑all approach apply to the extension
C
of stage‑discharge relationships at flow‑gauging weirs?
QQ  dvanced numerical modelling of the nonlinear
A
mechanical behaviour of a laterally loaded
pile embedded in stiff unsaturated clay
QQ  omparison of sinkhole size versus
C
borehole hazard classification
Volume 65 No 2 June 2023
ISSN 1021-2019 (print) | ISSN 2309-8775 (online)

PUBLISHER
South African Institution of Civil Engineering
Block 19, Thornhill Office Park,
Bekker Road, Vorna Valley, Midrand, South Africa
Private Bag X200, Halfway House, 1685, South Africa
Tel +27 11 805 5947, Fax +27 11 805 5971
http://www.saice.org.za, civilinfo@saice.org.za

CONTENTS
EDITOR-IN-CHIEF
Dr Peter Day
Consultant, Jones & Wagener (Pty) Ltd
Extraordinary Professor of Geotechnical Engineering
Stellenbosch University
day@jaws.co.za
CO-EDITOR-IN-CHIEF
Prof Gerhard Heymann 2 Spalling of concrete in fire –
University of Pretoria
gerhard.heymann@up.ac.za an overview and local relevance
ADMINISTRATIVE EDITOR
Verelene de Koker J E van der Merwe
Tel +27 11 805 5947, Mobile +27 83 378 3996
verelene@saice.org.za
SUB-EDITORS
Prof EP Kearsley University of Pretoria
10 Design criteria for upstream raised
Prof CP Roth
Prof A Visser
University of Pretoria
University of Pretoria
tailings storage facilities
Prof J Wium Stellenbosch University
Prof M Zuidgeest University of Cape Town J A Wates
BALANCE OF THE JOURNAL EDITORIAL PANEL
Prof Y Ballim University of the Witwatersrand
Prof W Burdzik Retired from the University of Pretoria
Prof N de Koker Stellenbosch University
17 Could a one-size-fits‑all approach apply
Prof J du Plessis Stellenbosch University
Prof GC Fanourakis University of Johannesburg
to the extension of stage‑discharge
Dr T Gaspar
Prof M Gohnert
Sheffield University
University of the Witwatersrand
relationships at flow‑gauging weirs?
Dr C Herold Umfula Wempilo Consulting
Prof A Ilemobade University of the Witwatersrand
O J Gericke, V H Williams
Prof SW Jacobsz University of Pretoria
Dr C MacRobert Stellenbosch University
Dr O Nnene University of Cape Town
Prof S Nwaubani University of the Witwatersrand
28 Advanced numerical modelling of the nonlinear
Prof W Steyn University of Pretoria
Dr J van der Merwe University of Pretoria
mechanical behaviour of a laterally loaded
Mr M van Dijk
Prof G van Zijl
University of Pretoria
Stellenbosch University
pile embedded in stiff unsaturated clay
Prof C Venter University of Pretoria
Dr E Vorster Zutari K T Braun, N Bakas, G Markou, S W Jacobsz
Prof R Walls Stellenbosch University
Prof A Zingoni University of Cape Town
PEER REVIEWING / OPEN ACCESS
The Journal of the South African Institution of Civil Engineering,
39 Comparison of sinkhole size versus
which is distributed internationally, is a peer-reviewed,
open-access journal licensed under a Creative Commons
borehole hazard classification
Attribution Licence (CC BY-NC-ND). Readers may therefore
freely use and share the content as long as they credit the
J Bunk
original creator and publisher, do not change the material
in any way, and do not use it commercially.

Creative Commons
44 Guidelines for the preparation of
Attribution Licence papers and technical notes
DESIGN AND REPRODUCTION
Marketing Support Services, Ashlea Gardens, Pretoria
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed
in this publication. Consequently, nobody connected with
the publication of this journal, in particular the proprietor,
the publisher and the editors, will be liable for any loss or
damage sustained by any reader as a result of his or her action
upon any statement or opinion published in this journal.
© South African Institution of Civil Engineering
and the authors 1
TECHNICAL PAPER
Spalling of concrete in fire –
an overview and
JOURNAL OF THE SOUTH AFRICAN
INSTITUTION OF CIVIL ENGINEERING

local relevance
ISSN 1021-2019 (print) | ISSN 2309-8775 (online)
Vol 65 No 2, June 2023, Pages 2–9, Paper 1410

DR JOHANN VAN DER MERWE (Pr Eng,

MSAICE) is a senior lecturer in the J E van der Merwe
Department of Civil Engineering at the
University of Pretoria. He completed his
BEng and MSc(Eng) degrees at Stellenbosch
University, and his doctorate at the Swiss Concrete has traditionally been considered as a material with favourable fire resistance.
Federal Institute of Technology in Zurich
The development of modern concretes with low permeability has, however, resulted in the
(ETH Zurich). He worked as a consulting
design engineer for a number of years before joining the University of increased occurrence of spalling when exposed to fire. Recent fire damage in South Africa
Pretoria in 2023. After first researching the material properties governing during the riots in July 2021 resulted in such damage.
the behaviour of concrete in fire and the seismic analysis of structures, he This paper presents an overview of explosive spalling of concrete in fire, recent local
is now concentrating his research efforts on topics related to improved
observations, and developments in the field. It is shown that various (often interrelated)
sustainability of building structures.
factors influence the susceptibility of concrete to spall in fire, and that much research is still
Contact details:
needed to fully understand the phenomenon and, specifically, how to accurately predict its
Department of Civil Engineering
University of Pretoria occurrence. Recent observations in South Africa support conclusions in the literature that, as an
Private Bag X20 accidental loading case or as life-safety performance criteria, overall structural integrity is rarely
Hatfield compromised. However, as the use of higher concrete grades increases, the relevance of this risk
Pretoria 0028
South Africa
should be considered by practising engineers.
E: vandermerwe.johann@up.ac.za

INTRODUCTION during a fire for normal concrete grades.

Fires in structures pose a significant As an accidental load case then, such dam-
threat to occupants and property owners. age might be considered as detrimental to
Structural engineers are responsible for structural integrity, but low hazard.
limiting the probability of injury, loss of Spalling of concrete is known to occur
life and damage to property to acceptable in some cases of fire exposure. This
levels. Traditionally, concrete has been con- typically leads to a loss of concrete cover,
sidered as incombustible, chemically stable, resulting in a loss of thermal insulation
and as a material with favourable insulat- to reinforcement and increased thermal
ing properties, performing well during a degradation of the confined concrete core
fire (Buchanan & Abu 2017). A gradual and high temperature exposure of rein-
decrease in compressive strength does, forcement steel. In severe cases, spalling
however, occur with increased concrete depth can extend into the confined core
material temperature. A marked reduction of reinforced concrete elements which,
in strength is associated with the 300°C together with high-temperature exposure
isotherm, which is often accompanied of reinforcement, can lead to collapse.
by a discolouration of the cement matrix Various types of fire-induced spalling
Keywords: fire exposure, pore pressure, explosive spalling, (Albrektsson et al 2011). Such thermal deg- have been defined to date, as summarised
reinforced concrete
radation can impair the continued service in Table 1. The earliest forms of spalling
List of symbols of reinforced concrete elements after a fire, are surface spalling, aggregate spalling,
b Biot’s coefficient (-)
but will seldom prevent safe evacuation and explosive spalling, of which explosive
E Young’s modulus of concrete (Pa)
ft tensile strength of concrete (Pa)
f t,20 tensile strength of concrete at ambient temperature (Pa) Table 1 Characteristics of the different types of fire-induced spalling (Khoury 2000)
f t,θ tensile strength of concrete at temperature θ (Pa)
k moisture content (%) Spalling type Time of occurrence Nature Sound Influence
patm atmospheric pressure (Pa)
px pore pressure at depth x (Pa) aggregate 7 – 30 min splitting popping superficial
Tf or θ fire (gas) temperature (°C)
Ts concrete surface temperature (°C) corner 30 – 90 min non-violent none can be serious
Tx concrete temperature at depth x (°C)
surface 7 – 30 min violent cracking can be serious
Wrupture rupture strain energy (J)
Wx strain energy in direction x (J)
explosive 7 – 30 min violent loud bang serious
εx strain in direction x (-)
ν Poisson’s ratio (-)
σℓ applied compressive stress (Pa)
σp pore pressure (Pa) Van der Merwe JE. Spalling of concrete in fire – an overview and local relevance.
J. S. Afr. Inst. Civ. Eng. 2023:65(2), Art. #1410, 8 pages. http://dx.doi.org/10.17159/2309-8775/2023/v65n2a1
σt thermal stress (Pa)
The Journal of the South African Institution of Civil Engineering, which is distributed internationally, is a peer-reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND). Readers may therefore freely use and share the content as long as they credit the original creator and publisher, do
2 not change the material in any way, and do not use it commercially. Copyright of this article remains with the authors.
 (a) (d)

(b)

spalling has occurred. Bailey (2002) con-

cluded that this is most likely due to slabs
acting in compressive membrane action
following thermal expansion, although
reduced loading benefits response as well.
The development of concrete with
increased strength and reduced permeability,
such as high-strength concrete (HSC) and
self-compacting concrete (SCC), holds many
advantages, such as increased durability and
reduced element sizes. These advantages
(c)
lead to its increased use in high-rise build-
ings, tunnel linings and warehouse floors.
However, the low permeability of such
compact modern concrete types leads to an
increased risk of explosive spalling during
fire exposure (Mindeguia et al 2013). This
makes the assessment of spalling susceptibil-
ity of concrete, the possible extent of damage
that might prevent safe evacuation during
a fire, ways of preventing its occurrence,
and the repair of such damage, increasingly
relevant to structural engineers. Recent local
observations illustrate this point.

Figure 1 E xamples of recent fire-induced spalling damage during a fire damage assessment by Recent observations of fire-induced
the author following the July 2021 riots in South Africa: (a) explosive spalling of the spalling of concrete in South Africa
steel-fibre-reinforced concrete floor, (b) spalling to the soffit and drop panel of an office The July 2021 riots in South Africa resulted
building, (c) spalling to the top of an exposed slab, and (d) corner spalling of a column in extensive fire damage to a warehouse
structure in Durban. In addition to par-
spalling is considered the most violent. Due Partial collapse of reinforced concrete tial collapse of the steel superstructure,
to the early, violent and often extensive building structures as a result of fire has extensive fire-induced spalling of concrete
nature of explosive spalling, it is considered been observed in the past. In many cases, elements resulted, as shown in Figure 1.
as the greatest hazard to reinforced con- however, reinforced concrete slabs are Severe explosive spalling of the steel-
crete structures if the extent is substantial. able to maintain structural integrity after fibre-reinforced concrete floor occurred in

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 3
isolated locations (Figure 1(a)) to depths of
up to half of the floor thickness. Similarly,
extensive spalling resulted to slab soffits
of the adjacent office building, exposing
bottom reinforcement (Figure 1(b)). Such
spalling around columns has been shown
to drastically reduce punching shear
resistance (Lu et al 2018). Localised fires
in reinforced concrete structures inside
the warehouse space showed substantial
spalling of the top of slabs, as well as cases
of corner spalling of columns, exposing
reinforcement to fire (Figures 1(c) and 1(d)).
PROPERTIES INFLUENCING
SPALLING
Numerous parameters influence the sus-
ceptibility of concrete to spall in fire, rang-
ing from material production to construc-
tion and operational conditions. Properties
can be grouped into the following four
categories (Van der Merwe 2019):
Concrete mixture composition
The aggregate type is said to influence spall-
ing susceptibility, with siliceous aggregate
concretes often reported as being more sus-
ceptible to spalling (Fib Bulletin 38: 2007).
Substantial thermal damage can result due
to the difference in the thermal expansion
coefficient of aggregate and the surrounding
cement paste. Quartz in siliceous aggregates
is generally associated with a phase change
at high temperatures that results in a rapid
increase in volume and subsequent thermal
damage to concrete (Bažant & Kaplan 1996).
Although some contradictory results
have been reported in the literature, there
appears to be general agreement that a
larger coarse aggregate size reduces spalling
susceptibility. Experimental results by Pan et
al (2012) support this, suggesting the cause
to be related to an increased characteristic
length of the fracture process zone that is
associated with larger aggregates, although
increased permeability is likely an additional
contributing factor (Van der Merwe 2019).
High cement and silica fume content is
also said to increase spalling susceptibility,
correlating with lower permeability and a
subsequent higher internal pressure build-
up potential (e.g. Klingsch 2014; Kodur &
McGrath 2006).
Material properties of
hardened concrete
It has often been stated in the literature that
greater compressive strength is often associ-
ated with increased spalling susceptibility
4 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
(e.g. Majorana et al 2010). General consensus
is that this is due to the compact microstruc-
ture that is associated with greater compres-
sive strength. A strong correlation has been
shown to exist between the compressive
strength of concrete and its intrinsic permea-
bility (Van der Merwe 2022c), suggesting that
higher concrete grades can be expected to be
associated with higher pore pressure build-up
during fire exposure. Related to such a com-
pact microstructure is the permeability of
concrete, which governs pore pressure build-
up potential (Bažant & Kaplan 1996). Since
the permeability of concrete can be reduced
without increasing compressive strength,
Hertz (2003) concluded that permeability is a
more relevant indication of spalling risk than
compressive strength.
In addition to permeability, the moisture
content determines pore pressure build-up
during fire exposure, and is therefore consid-
ered by many as a crucial parameter related
to spalling susceptibility (e.g. Hertz 2003).
In addition to its influence in pore pressure
build-up, increased moisture content can be
expected to increase the thermal stress gradi-
ent (Fib Bulletin 38: 2007). Latent heat during
the phase change from liquid water to vapour
is associated with an increased specific heat
of concrete, with a subsequent slower increase
in temperature away from the heated surface
and a resulting increase in thermal stress.
Similarly, thermal properties that affect ther-
mal stress development are also associated
with spalling risk, including the specific heat
and thermal conductivity of concrete.
Since fracture energy determines the
specific energy required for crack propaga-
tion, lower fracture energy is associated with
increased spalling risk. Bažant and Prat (1988)
concluded from experimental results that the
fracture energy of concrete gradually decreas-
es with increased temperature, thereby mak-
ing concrete gradually more susceptible to
spalling with increased temperature.
Structural properties
Previous investigations have shown that the
spalling susceptibility of concrete generally
increases (i) with increased element size
(Boström et al 2007), (ii) where structural
geometry results in stress concentration at
sharp corners (Khoury & Anderberg 2000),
and (iii) where thermal restraint prevents
expansion of concrete during heating
(Majorana et al 2010).
Loading
It is generally accepted that an increased
heating rate increases spalling susceptibility
due to increased thermal stress development
(Kodur & Phan 2007). Experimental obser-
vations have also shown that moderate com-
pressive stress increases spalling susceptibil-
ity (Carré et al 2013). Moderate compressive
stress adds to the overall internal stress
state, but has also been shown to reduce the
intrinsic permeability of concrete (Van der
Merwe 2022b), which in turn can result in
increased pore pressure development.
PLAUSIBLE MECHANISMS
OF SPALLING
Various mechanisms for explosive
spalling have been proposed, and some
academic debate in this regard continues
to date. Shorter and Harmathy (1961) and
Harmathy (1965) proposed that pore pres-
sure build-up in the concrete matrix at ele-
vated temperatures is the main mechanism
responsible for spalling. Once moisture
evaporation starts, the resulting pressure
gradient forces part of the moisture to the
heated surface, whilst the direction of heat
flux forces the remainder to the cooler inner
region. Here condensation is said to result in
a region of increased moisture content. As
heating continues, two distinct zones result
– a dry region near the heated surface, and a
saturated zone (moisture clog) deeper in the
concrete section (Figure 2). The low perme-
ability in this moisture clog results in gradu-
ally increasing pore pressure and finally in
spalling when the high-temperature tensile
strength of concrete is exceeded.
Despite such a moisture clog having been
experimentally shown to develop, it is widely
accepted that pore pressure alone cannot be
responsible for spalling since the mechanism
does not account for increased spalling risk
due to confinement (Jansson & Boström
2012) and since resulting pore pressure is
usually well below typical ranges of concrete
tensile strength (Bažant & Thonguthai 1979).
The high heating rate imposed during
fire exposure results in a high temperature
gradient over the depth of a concrete sec-
tion which, in turn, leads to the develop-
ment of thermal stress. The compressive
stress state that results near the heated
surface was proposed by Saito (1966) as
a mechanism of explosive spalling. It was
suggested that such compressive stress
induces cracks that separate a surface
layer. Thermal buckling of this weakened
delaminated layer is the proposed cause of
spalling. Experimental observations have
shown, however, that spalling can occur in
some concretes at heating rates much lower
 Pressure difference Heat flux

Px
Tf
(1)

Ts (2)

σt (i)

(ii)
Patm Patm σℓ

Tx
(iii)
σp

Figure 3 T he combined stress state mechanism

Dry zone
Figure 2 Pore pressure spalling mechanism proposed by Harmathy (adapted from Harmathy 1965)
that those that would impose meaningful
thermal stress, suggesting that thermal
stress alone cannot be a mechanism of
spalling (Klingsch 2014).
The most widely accepted mechanism
is the combined stress state mechanism
proposed by Zhukov (1976). With reference
to Figure 3, this mechanism suggests that
explosive spalling occurs when the combined
effect of thermal stress, pore pressure devel-
opment and level of applied compressive
stress exceeds the temperature-dependent
tensile strength of concrete.
1.2
1.0
Saturated zone
Considering the total strain in the
direction of spalling (Equation 1), pore
pressure (σ p) acts directly in this direction,
whilst Poisson effects from thermal stress
(σ t) and applied compressive stress (σℓ) add
to this effect.
1 1
єx = ∙ [σ p + v ∙ (σℓ + 2 ∙ σ t)](1)
E 2∙E
Explosive spalling is said to occur when the
strain energy (Wx, Equation 2) exceeds the
rupture strain energy (Wrupture, Equation 3).
suggests that spalling of concrete
(region 1) occurs when the concrete
tensile strength is exceeded (at depth
shown by 2) and a crack opens due
to a stress state resulting from the
combined effect of (i) thermal stress
(σt), (ii) applied compressive stress (σℓ)
and (iii) pore pressure (σp) (adapted
from Zhukov 1976)
1 σ
Wx = ∙ єx ∙ σx = p ∙ [σp + v ∙ (σℓ + 2 ∙ σt)](2)
2 2∙E
Wrupture = ∙ f t2(3)
Combining Equations 2 and 3, it follows
that explosive spalling is initiated when the
combined internal stress state exceeds the
tensile strength of concrete, as shown in
Equation 4. With increased material tem-
perature, the tensile strength of concrete
can be expected to gradually decrease as
shown in Figure 4.

f t2 = σ p2 + v ∙ σ p ∙ (σℓ + 2 ∙ σ t)(4)
0.8
f t,θ/f t,20 (–)

0.6 PREDICTIVE MODELS

Numerous attempts have been made to
develop a means of predicting the occur-
0.4
rence of explosive spalling of concrete in fire,
ranging from nomograms, analytical models
0.2 and numerical models. To date, however,
none of these have been validated as a reli-
able means of predicting this phenomenon.
0
0 100 200 300 400 500 600 700 800
θ (°C) Moisture clog theory
Harmathy (1965) proposed analytical
Figure 4 T ensile strength of concrete has been shown to gradually decrease with increasing expressions for the movement of a mois-
temperature (adapted from Van der Merwe 2022a); within the temperature range ture clog in concrete. Based on these
typically associated with spalling (Fib Bulletin 46: 2008), a tensile strength reduction of expressions, he proposed a limit criterion
up to 50% can be expected for the pore saturation associated with

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 5
 and the thickness of a concrete element
1.0 heated from two sides (Figure 6). An
adapted version of this nomogram was later
incorporated in Eurocode (EN 1995).
0.8
Numerical models
Degree of saturation (–)

Each of the two nomograms described above

0.6
0.4
0.2
0
1E-13
Figure 5 N
 omograms proposed by Harmathy (1965) and Sertmehemetoglu (1977) for spalling
risk considering concrete permeability and the degree of saturation (adapted from
Harmathy 1965 and Sertmehemetoglu 1977)
fire-induced spalling. His proposed nomo-
gram (Figure 5) provides a graphical means
of evaluating the risk of explosive spalling
based on a limit value for pore saturation
with permeability. This model was modi-
fied by Sertmehemetoglu (1977) to account
for the difference in permeability, viscosity,
and pressure between the moisture clog
and the heated surface. Comparing these
two models in Figure 5, shows that the
modified model suggests a narrower range
clearly accounts for parameters known to
be associated with explosive spalling, but
neglects other known contributing param-
Moisture eters. Numerical models have been proposed
clog spalling
to account for all known effects that contrib-
ute to explosive spalling. Gawin et al (1999)
No spalling
proposed a hygro-thermal model which,
amongst others, accounts for the effect of
moisture movement and heat flux. Their
model for the high-temperature development
1E-12 1E-11 1E-10 1E-09 1E-08 1E-07 1E-06 of intrinsic permeability was subsequently
Permeability (m2) considered by Dwaikat and Kodur (2009) to
Harmathy (1965) Sertmehemetoglu (1977) propose a hydrothermal model for the assess-
ment of spalling risk. The model accounts
for temperature variation, pore pressure and
moisture movement that are considered as
coupled equations to be solved numerically.
Spalling is then identified when the pore
of permeability associated with spalling pressure exceeds the high-temperature
susceptibility, which can be expected to tensile strength of concrete. Elaborating
shift to higher or lower permeability values on this model to account for pore pressure
depending on the concrete tensile strength. in all directions, and for Biot’s coefficient,
which determines the extent to which pore
Level of compressive stress pressure affects the internal stress state of
and element thickness the concrete matrix, Lu (2015) proposed the
Based on experimental results by Meyer- numerical model shown in Equation 5.
Ottens (1972), Sertmehemetoglu (1977)
proposed a nomogram for spalling suscep- √(1 – 2v) ∙ b2 ∙σp2 + v ∙ (b ∙ σp) ∙ (σℓ + 2 ∙ σt) ≥ f t
tibility based on the level of applied stress (5)

20 35
Explosive spalling
Surface spalling
No spalling 30

15
25
Compressive stress (MPa)

(1)
20
σ (N/m2)

10
Spalling risk
15
(2)

10
5 Spalling
unlikely
5

0 0
0 40 80 120 160 0 50 100 150 200
(a) Specimen thickness (mm) (b) t (mm)

Figure 6 T he nomogram for spalling risk proposed by Sertmehemetoglu (adapted from Sertmehemetoglu 1977) proposed a means of assessing spalling
risk based on the thickness of a concrete element and the level of applied load; an adapted version of this nomogram was included in an earlier
version of Eurocode (adapted from EN 1995)

6 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
Considering a 150 mm thick concrete slab
exposed to fire on one side, and with a refer-
ence concrete tensile strength of 2.5 MPa, Lu
(2015) proposed a boundary intrinsic perme-
ability value of 2 × 10–17 m2. The correlation
between concrete tensile strength and
intrinsic permeability investigated by Van
der Merwe (2019) suggests that such intrinsic
permeability values are, however, unlikely for
the concrete tensile strength assumption.
GENERAL METHODS
OF PREVENTION
Experimental observations of explosive
spalling have shown the occurrence of this
phenomenon to be complex and often errat-
ic. Fib Bulletin 38 (2007) reports a test series
of ten similar specimens where five samples
spalled, whilst five did not. Elsewhere this
observation is supported, mentioning that,
for concrete samples from the same batch
and under identical curing and testing
conditions, some samples would spall whilst
others would not (Connolly 1995; Majorana
et al 2010). To this extent, Bisby et al (2014)
recommend that “… attempts should be
made to avoid spalling … in practice, since
spalling introduces uncertainties that are
very difficult to rationally account for in
design.” Where a risk of spalling is therefore
suspected, or where the consequence of
explosive spalling is expected to have a sub-
stantial effect on structural integrity or the
safe evacuation of occupants, preventative
methods should be considered.
Addition of fibres
In addition to improving ductility, the
addition of steel fibres to concrete increases
its tensile strength. Since it is gener-
ally believed that spalling occurs when the
(a)
Figure 7 H
 SC slabs after fire testing: (a) without a protective lining, explosive spalling started after 15 minutes and continued gradually to a depth
of 100 mm after stopping the test, (b) where a 10 mm protective lining was provided, spalling occurred with a single load explosion after
119 minutes to the same depth (Klingsch et al 2013 – ETH Zurich, open-access source for non-commercial use)
Journal of the South African Institution of Civil Engineering
high-temperature stress state in concrete
exceeds its tensile strength, steel fibres can
be expected to increase the resistance to
explosive spalling. Nevertheless, contra-
dicting experimental observations of its
beneficial effect have been reported, with
some tests showing improved spalling resist-
ance (Kodur et al 2003), whilst others notice
no substantial improvement (Klingsch et al
2013). Moreover, the addition of steel fibres
has been noted to increase the violent nature
of explosive spalling in some cases due to the
release of a greater build-up of energy accom-
modated by the increased tensile strength
(Fib Bulletin 38: 2007). This supports recent
observations of a steel-fibre-reinforced con-
crete warehouse floor (Figure 1a).
It is widely agreed that the addition of
polypropylene fibres (PP-fibres) reduces
the risk of spalling. Explosive spalling of
concrete with compressive strength of up to
150 MPa has been found to be avoided when
adding 0.05 – 0.10% PP-fibres (Fib Bulletin
38: 2007). Most conclusions in literature
seem to agree that the observed benefit of
adding PP-fibres is related to increased per-
meability and improved moisture transport,
thereby reducing pore pressure build-up (e.g.
Heo et al 2012; Klingsch et al 2013; Maluk
et al 2017). Increased entrained air during
the concrete mixing process, and poor
adhesion between fibres and the surround-
ing concrete matrix have been proposed
as potential mechanisms for the beneficial
effect of adding PP-fibres, but the formation
of continuous channels after fibres melt, and
differential thermal expansion are the two
most widely argued mechanisms. Following
experimental tests of HSC with PP-fibres,
Kalifa et al (2001) studied test samples with
a scanning electron microscope. They con-
cluded that molten PP-fibres are absorbed
(b)
Volume 65 Number 2 June 2023
into the surrounding porous cement matrix.
The interconnected channels created in this
manner are said to increase permeability
and, subsequently, reduce pore pressure.
Another proposed mechanism suggests that
there exists a difference in thermal expan-
sion between PP-fibres and the surrounding
concrete matrix, resulting in micro cracks
that increase overall permeability and,
subsequently, reduce pore pressure. Zhang
et al (2018) investigated ultra-high perfor-
mance concrete with PP-fibres after heating
to elevated temperatures at 5°C/min. A
subsequent review of samples with a scan-
ning electron microscope revealed micro
cracks in the concrete matrix between fibres
prior to melting, suggesting that the thermal
expansion mismatch between fibres and the
cement matrix is responsible for the reduced
spalling risk. Despite the variation in
conclusions, it is clear that PP-fibres reduce
the risk of explosive spalling by increasing
permeability. It is argued here that the
interconnectedness of increased porosity
that is required to increase permeability, is
more likely to result from the formation of
microcracks than absorbed molten fibres.
Protective barriers
The aim of protective barriers, be they
panels, mortar or intumescent paint, is to
minimise heat transfer to concrete in the
event of a fire. No design criteria are cur-
rently available for the use of these measures
to avoid explosive spalling. As such, care-
ful experimental validation of protective
barriers is typically required. Reviewing a
series of fire tests by Klingsch et al (2013),
Lu (2015) concluded that the provision
of an insufficient protection layer can, in
some cases, increase the violent nature
of explosive spalling. Figure 7 compares
7
explosive spalling of a slab with and without
a protective lining (Klingsch et al 2013).
Although the protective lining delayed the
occurrence of spalling, damage occurred to
the same depth. Moreover, compared to the
gradual, layer-by-layer spalling observed for
the slab without a protective layer, spalling
occurred with a single loud explosion where
the protective layer had been applied.
Reinforcement details
A series of experimental and numerical
investigations of the behaviour of reinforced
concrete columns in fire were conducted by
Kodur and McGrath (2006), and Kodur and
Phan (2007). They concluded that transverse
confinement reinforcement details that
are typically recommended in regions of
seismic hazard are well suited to improve
the explosive spalling resistance of columns.
The detail calls for (i) a reduced transverse
reinforcement diameter with decreased
vertical spacing instead of larger diameters
at greater spacing, (ii) the provision of cross
ties, and (iii) anchorage of tie ends into con-
fined concrete, i.e. a 135° hook instead of the
more common 90° hook (Figure 8).
CURRENT REGULATIONS
AND GUIDELINES
Eurocode 2 (EN 2004) addresses the fire
design of concrete structures and is per-
haps the most suitable current reference,
since the suite of SANS standards does
not include similar clauses. This standard
recommends that the occurrence of spalling
should either be prevented, or that the
engineer should consider the consequences
of spalling on the structural capacity of ele-
ments. Spalling is addressed separately for
normal strength concrete (NSC) and high
strength concrete (HSC). It is suggested
that spalling of NSC is unlikely to occur if
the initial free moisture content is less than
3%. An earlier version of this standard (EN
1995) mentioned that this moisture content
would most likely not be exceeded if a con-
crete element is subjected to internal expo-
sure. Where this limit is exceeded, the risk
of spalling should be more carefully investi-
gated, considering (i) the moisture content,
(ii) aggregate type, (iii) concrete permeabil-
ity, and (iv) the expected (or design) heating
rate. It does, however, give no guidance as
to how these factors should be reviewed to
assess spalling susceptibility. Nevertheless,
the standard recommends that no further
in-depth spalling assessment is necessary
if the tabulated concrete material property
8 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
(1)
(2)
Figure 8 E xperimental observations by
Kodur and McGrath (adapted from
Kodur & McGrath 2007) suggest that
improved confinement detailing
of stirrups to columns improves
the spalling resistance of columns:
(1) ends of stirrups should be
anchored into the confined core, and
(2) cross ties should be provided
data and the recommended reinforcement
details are considered by the engineer.
HSC up to class C80/95 with a silica fume
content of less than 6% of the cement mass, is
treated as NSC. For higher concrete classes,
or where the silica fume content exceeds 6%,
four possible methods are recommended to
prevent the occurrence of spalling:
(A) providing mesh reinforcement near the
exposed concrete surface
(B) prescribing concrete that has been
proven (experimentally or by experi-
ence) not to be susceptible to spalling
(C) providing a protective layer that has
been shown to prevent spalling
(D) including a minimum of 2 kg/m3
monofilament polypropylene fibres.
However, even when adhering to regulations,
spalling can still occur. A full-scale fire test
was performed on a seven-storey reinforced
concrete building in Cardington in 2001
(Bailey 2002). Adhering to prescriptive rules,
Eurocode recommendations suggest that
spalling should not occur and that its occur-
rence can be ignored. Nevertheless, severe
spalling did occur. The outcome highlights
our poor understanding of the phenomenon
of explosive spalling, and the limited means
available to practising engineers to assess the
spalling susceptibility of concrete.
DEVELOPMENTS
Recent findings are set to influence the
evolution of Eurocode provisions pertain-
ing to the susceptibility of concrete to
spall explosively in fire. Eurocode (EN
2004) mentions that spalling should not
occur if the moisture content is below
k% by weight, with 2.5 ≤ k ≤ 3.0% and a
recommended value of 3%. Although the
moisture content of concrete is undoubt-
edly an important parameter affecting the
propensity of concrete to spall during fire
exposure, it is difficult to predict what the
moisture content of a concrete element will
be, and equally hard to influence its value.
Moreover, spalling has been observed at
lower moisture content levels (e.g. Zheng et
al 2010). Results from various studies suggest
that the relevance of moisture content as a
parameter affecting spalling susceptibility
depends on the type of aggregate, cross-
sectional shape of the element, level of com-
pressive stress, concrete mixture and heating
rate (Jansson 2013). It has been suggested that
such a moisture content limit be replaced by
a list of concrete conditions known to result
in a high moisture content and limited drying
(Jansson 2013). These can include concrete
exposed to saturated conditions or with per-
manent formwork that prevents drying.
CONCLUDING REMARKS
Although less severe types of fire-induced
spalling might affect normal concrete
strength grades to a limited extent, the
susceptibility of concrete to spall explosively
in fire is likely to become an increasingly
relevant risk to be considered by structural
engineers as the use of denser concretes and
higher concrete grades increase in South
Africa. This paper highlighted some key
considerations, but also showed that, to a
large extent, this phenomenon remains poorly
understood. Much research is still needed
in order to more accurately assess this risk.
To date, the only accurate method of assess-
ing the susceptibility of concrete to spall
explosively in fire, is to perform fire tests of
representative concrete elements. Since both
the size and geometry of a concrete element
influence the risk of spalling, small-scale
samples, such as standard concrete cubes
or cylinders, should not be considered as
representative (e.g. Boström et al 2007). Due
to the great cost of such fire tests, various
studies attempted to develop a numerical
means of assessing spalling susceptibility.
Significant progress has been made towards
such a numerical means of assessment (Lu
2015; Lottman 2017), but fundamental input
parameters for such models have not been the
focus of many studies (Van der Merwe 2019).
Considering results reported to date, the most
effective method of prevention is to include
polypropylene fibres to a concrete mixture.
ACKNOWLEDGEMENTS
The author wishes to thank Prof Dr Mario
Fontana, retired professor at ETH Zurich,
for his guidance during the author’s doctoral
study focusing on material properties related
to explosive spalling of concrete in fire.
REFERENCES
Albrektsson, J, Flansbjer, M, Lindqvist, J E & Jansson, R
2011. Assessment of concrete structures after fire. SP
Report 2011:19. SP Technical Research Institute of
Sweden.
Bailey, C 2002. Holistic behaviour of concrete buildings
in fire. Proceedings of the Institution of Civil
Engineers, Structures and Buildings, 3: 199–212.
Bažant, Z P & Thonguthai, W 1979. Pore pressure
in heated concrete walls: Theoretical prediction.
Magazine of Concrete Research, 31(107): 67–76.
Bažant, Z P & Prat, P C 1988. Effect of temperature
and humidity on fracture energy of concrete. ACI
Materials Journal, 85(4): 262–271.
Bažant, Z P & Kaplan, M F 1996. Concrete at
High Temperatures: Material Properties and
Mathematical Models. New York: Longman.
Bisby, L, Mostafaei, H & Pimienta, P 2014. State-of-
the-art on fire resistance of concrete structures:
structure-fire model validation. Draft report
submitted to Applied Research Associates,
Albuquerque, NM.
Boström, L, Wickström, U & Adl-Zarrabi, B 2007. Effect
of specimen size and loading conditions on spalling
of concrete. Fire and Materials, 31: 173–186.
Buchanan, A H & Abu, A K 2017. Structural design for
fire safety. Hoboken, NJ: Wiley.
Carré, H, Pimienta, P, La Borderie, C, Pereira, F &
Mindeguia, J-C 2013. Effect of compressive loading
on the risk of spalling. MATEC Web of Conferences,
6: 01007.
Connolly, R J 1995. The spalling of concrete in fires. PhD
Thesis. Birmingham, UK: Aston University.
Dwaikat, M B & Kodur, V 2009. Hydrothermal model
for predicting fire-induced spalling in concrete
structure systems. Fire Safety Journal, 44: 425–434.
EN (European Standard) 1995. ENV 1992-1-2 (1995).
Eurocode 2: Design of concrete structures. Part 1-2:
General rules – Structural fire design. Brussels:
European Committee for Standardization.
EN 2004. EN 1992-1-2 (2004). Eurocode 2: Design
of concrete structures. Part 1-2: General rules –
Structural fire design. Brussels: European Committee
for Standardization.
Fib Bulletin 38: 2007. State-of-the-art report. Fire design
of concrete structures. Materials, structures and
modelling. Task Group 4.3-1. Lausanne, Switzerland:
International Federation for Structural Concrete.
Fib Bulletin 46: 2008. State-of-the-art report. Fire design
of concrete structures. Structural behaviour and
Journal of the South African Institution of Civil Engineering
assessment. Task Group 4.3. Lausanne, Switzerland:
International Federation for Structural Concrete.
Gawin D, Majorana, C E & Schrefler, B A 1999.
Numerical analysis of hygrothermal behaviour and
damage of concrete at high temperature. Mechanics
of Cohesive-Frictional Materials, 4: 37–74.
Harmathy, T Z 1965. Effect of moisture on the fire
endurance of building elements. In: Moisture
in Materials in Relation to Fire Tests. West
Conshohocken, PA: ASTM International, 74–95.
Heo, Y-S, Sanjayan, J G, Han, C-G & Han, M-C 2012.
Limited effect of diameter of fibres on spalling
protection of concrete in fire. Materials and
Structures, 45: 325–335.
Hertz, K D 2003. Limits of spalling of fire-exposed
concrete. Fire Safety Journal, 38: 103–116.
Jansson, R & Boström, L 2012. Determination of fire
spalling of concrete: Relevance of different test
methods. Proceedings, 7th International Conference
on Structures in Fire, Zurich, Switzerland, 6–8 June
2012, pp 581–587.
Jansson, R 2013. Fire spalling of concrete: Theoretical
and experimental studies. PhD Thesis Stockholm,
Sweden: KTH Royal Institute of Technology.
Kalifa, P, Chéné, G & Gallé, C 2001. High-temperature
behaviour of HPC with polypropylene fibres: From
spalling to microstructure. Cement and Concrete
Research, 31: 1487–1499.
Khoury, G A 2000. Effect of fire on concrete and
concrete structures. Proceedings of Structural
Engineering Materials, 2: 429–447.
Khoury, G A & Anderberg, Y 2000. Concrete spalling
review. Fire Safety Design, 60: 5–12.
Klingsch, E, Frangi, A & Fontana, M 2013. Explosive
spalling of concrete in fire. IBK Test Report 352.
Zurich, Switzerland: ETH Institute of Structural
Engineering (IBK)
Klingsch, E W H 2014. Explosive spalling of concrete in
fire. PhD Thesis. ETH Zurich, Switzerland.
Kodur, V K R, Cheng, F-P, Wang, T-C & Sultan, M-A
2003. Effect of strength and fiber reinforcement on
fire resistance of high-strength concrete columns.
Journal of Structural Engineering, 129(2): 253–259.
Kodur, V K R & McGrath, R 2006. Effect of silica fume
and lateral confinement on fire endurance of high
strength concrete columns. Canadian Journal of Civil
Engineering, 33: 93–102.
Kodur, V K R & Phan, L 2007. Critical factors governing
the fire performance of high-strength concrete
systems. Fire Safety Journal, 42: 482–488.
Lottman, B B G 2017. The spalling mechanism of fire
exposed concrete. PhD Thesis. Technical University
of Delft, Netherlands.
Lu, F2015. On the prediction of concrete spalling under
fire. PhD Thesis. ETH Zurich, Switzerland.
Lu, F, Beartschi, R, Van der Merwe, J & Fontana, M
2018. The effect of explosive spalling on punching
shear strength of concrete slabs exposed to fire.
Volume 65 Number 2 June 2023
Proceedings, 10th International Conference on
Structures in Fire. Ulster University, Belfast, Ireland,
pp 77–84.
Majorana, C E, Salomoni, V A, Mazzucco, G &
Khoury, G A 2010. An approach for modelling
concrete spalling in finite strains. Mathematics and
Computers in Simulation, 80: 1694–1712.
Maluk, C, Bisby, L & Terrasi, G P 2017. Effects of
polypropylene fibre type and dose on the propensity
for heat-induced concrete spalling. Engineering
Structures, 141: 584–595.
Meyer-Ottens, C 1972. Zur Frage der Abplatzungen
an Betonbauteilen aus Normalbeton bei
Brandbeanspruchung. PhD Thesis. Braunschweig
University of Technology, Germany.
Mindequia, J-C, Pimienta, P, Carré, H & La Borderie, C
2013. Experimental analysis of concrete spalling due
to fire exposure. European Journal of Environmental
and Civil Engineering, 17(6): 453–466.
Pan, Z, Sanjayan, J & Kong, D 2012. Effect of aggregate
size on spalling of geopolymer and Portland cement
concretes subjected to elevated temperatures.
Construction and Building Materials, 36: 365–372.
Saito, H 1966. Explosive spalling of prestressed concrete
in fire. Bulletin of the Fire Prevention Society of Japan,
15(2): 23–30.
Sertmehemetoglu, Y 1977. On a mechanism of spalling of
concrete under fire conditions. PhD Thesis. University
of London, UK.
Shorter, G W & Harmathy, T Z 1961. Discussion: “The
fire-resistance of prestressed concrete beams” by L A
Ashton & S C C Bate. Proceedings of the Institute of
Civil Engineers, 21(2): 313–315.
Van der Merwe, J E 2019. Constitutive models towards
the assessment of concrete spalling in fire. PhD Thesis.
ETH Zurich, Switzerland.
Van der Merwe, J E 2022a. Evaluation of concrete tensile
strength as a function of temperature. Construction
and Building Materials, 329: 127179.
Van der Merwe, J E 2022b. Effect of compressive stress
on the intrinsic permeability of concrete. Results in
Materials, 14: 100280.
Van der Merwe, J E 2022c. Towards the estimation of
concrete intrinsic permeability. Cement and Concrete
Research, 159: 106885.
Zhang, D, Dasari, A & Tan, K H 2018. On the
mechanism of prevention of explosive spalling in
ultra-high performance concrete with polypropylene
fibers. Cement and Concrete Research, 113: 169–177.
Zheng, W Z, Hou, X M, Shi, D S & Xu, M X 2010.
Experimental study on concrete spalling in
prestressed slabs subjected to fire. Fire Safety Journal,
45: 283–297.
Zhukov, V 1976. Reasons of explosive spalling of
concrete by fire. Beton I zhelezobeton (Concrete and
Reinforced Concrete), Vol 3.
9
TECHNICAL PAPER
Design criteria for
upstream raised tailings
JOURNAL OF THE SOUTH AFRICAN
INSTITUTION OF CIVIL ENGINEERING

storage facilities
ISSN 1021-2019 (print) | ISSN 2309-8775 (online)
Vol 65 No 2, June 2023, Pages 10–16, Paper 1501

JOHN WATES (Pr Eng, FSAICE) is currently

non-executive chair of Fraser Alexander (a J A Wates
mining services company) and provides
independent consulting services through
John Wates Consulting. He has over 45 years
of experience in design, construction and For more than 100 years upstreaming has been the dominant raising method for tailings
operation of mine tailings storage facilities,
storage facilities in Southern Africa. However, there is a growing view worldwide that
water dams and industrial and domestic
waste disposal sites. He serves, or has served, on more than 20 upstreaming is inherently less safe than centreline or downstream raising of tailings storage
independent tailings and water dam review boards, and has been facilities. Considering that more than 90% of the 250 plus operational tailings storing facilities
appointed as the Approved Professional Person (APP) for more than 50 in Southern Africa are upstream raised and cannot economically be converted to, or replaced
water and tailings dams up to Category III in South Africa. He has also
by, downstream raised facilities, the arguments for and against upstreaming were examined
participated in several failure investigations, the most notable of which
was the Cadia TSF failure in New South Wales, Australia. in this study. It was found that no compelling reason exists to ban upstreaming in Southern
Africa, provided that failure mechanisms are thoroughly considered to prevent serious design
Contact details:
John Wates Consulting omissions. The ALARP (as low as reasonably practicable) approach is introduced as a practical
PO Box 1634 method to deal with the risk of failure, and its hierarchical application is illustrated.
Rivonia 2128
South Africa
T: +27 11 231 3900
E: johnwates@jwcons.co.za INTRODUCTION Morgenstern (2018) stated in his De
Upstreaming has been the dominant rais- Mello lecture: “At this time, there is a crisis
ing method for tailings storage facilities in associated with concern over the safety
Southern Africa for more than 100 years. of tailings dams and lack of trust in their
Many of the facilities were positioned in design and performance. This crisis has
the densely populated towns and cities that resulted from recent high-profile failures
were built around the Witwatersrand gold of dams at locations with strong technical
reefs and the eastern and western limbs experience, conscientious operators and
of the platinum belt. Since 1886, when established regulatory procedures.” It is
gold was discovered in South Africa, there therefore indisputable that we must side
have been several concerning incidents with initiatives to regain trust. To do so
and two significant failures which claimed we need to focus on changing those things
lives, namely the Bafokeng failure in 1974 that will make a difference and not on
(Jennings 1979) and the Merriespruit those that will not.
failure in 1994 (Wagener 1997). Since the Banning of upstreaming is in most
introduction of the South African National instances promoted by practitioners from
Standard (SANS) 10286 code of practice the western coastal countries of South and
(SANS 1998), which focuses on improving North America where seismic activity is
management of tailings storage facilities, intense. In regions with extreme seismicity,
there have been no further failures of such as Chile, Peru and British Columbia,
significance. (The Jagersfontein failure the ban makes sense, but the reasons are
occurred since writing this paper and has not that clear for Brazil, where upstreaming
therefore not been considered). In the same has been banned following the sequence
period, upstreaming has been banned in at of failures of their upstream dams ending
least two countries where seismicity is low, with the Brumadinho failure which had
and probably not strong enough to pose a catastrophic consequences. Banning of
risk to tailings facilities. Hence there is a upstreaming in the seismically active regions
growing view worldwide that upstreaming on the eastern extremes of the Americas
is inherently less safe than centreline or in the 1970s has been shown to have been
downstream raising of tailings storage prudent, but it remains to be seen whether
facilities. Some are going as far as suggest- the ban in Brazil will deliver improved trust
ing that the practice should be banned based on a reduced incidence of failures.
across the globe. Although the Brazilian regulators thought so

Wates J. Design criteria for upstream raised tailings storage facilities.

J. S. Afr. Inst. Civ. Eng. 2023:65(2), Art. #1501, 7 pages. http://dx.doi.org/10.17159/2309-8775/2023/v65n2a2
Keywords: tailings, upstreaming, risk, ALARP, deposition
The Journal of the South African Institution of Civil Engineering, which is distributed internationally, is a peer-reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND). Readers may therefore freely use and share the content as long as they credit the original creator and publisher, do
10 not change the material in any way, and do not use it commercially. Copyright of this article remains with the authors.
at the time, it is not clear whether it is both
necessary and sufficient to ban upstreaming
to regain public trust. It is also possible that
banning of upstreaming is taking attention
away from the more important interventions
such as improved governance and mitigation
of weak engineering for all types of facilities
that should be implemented by the mining
industry, including its consultants, as a mat-
ter of urgency.
Morgenstern (2018) in his lecture went
further to say: “I side with the views of
Martin and McRoberts (1999), and others
before them (Lenhart 1950; Vick 1992),
that there is nothing wrong with upstream
tailings dams, provided that key principles
are adhered to in the design, construction
and operation of such dams.” Boswell and
Sobkowicz (2018) corroborated this view
by stating that, “For every failed tailings
dam, there exists a large number of stable,
well managed tailings facilities around
the world. These structures do not achieve
stability by chance, coincidence or serendip-
ity. Their stability is the result of sustained
effort over many years by the engineers that
design and build them …”
Where does that leave the Southern
African mining industry? Should Southern
Africa follow suit and eliminate upstream-
ing? More than 90% of the 250 plus opera-
tional tailings storage facilities (TSFs) in
Southern Africa are upstream raised and
cannot economically be converted to, or
replaced by, downstream raised facilities.
The local industry is therefore reluctant
to follow the call to move away from
upstreaming. Although there are signifi-
cant consequences for the industry which
is substantially invested in upstreaming,
the argument against elimination of
upstreaming for economic reasons is not
sufficient on its own, since safety must take
priority. In this regard, the author sides
with the views of Martin and McRoberts
(1999) and McRoberts et al (2017), and
with Morgenstern’s (2018) conclusion that
upstreaming is not inherently flawed, pro-
vided that key principles are observed. This
paper therefore serves to examine where
we have gone wrong with our upstreaming
practices and what needs to be done to
ensure that we can build upstream facilities
in a safe and sustainable way in future.
ARGUMENTS FOR AND
AGAINST UPSTREAM RAISING
There are good arguments both for and
against upstreaming. The arguments are,
Journal of the South African Institution of Civil Engineering
however, not always placed in context,
since those who are against may well be
correct that upstreaming cannot be safely
done in the regions with which they are
familiar, for legitimate reasons. Some of the
arguments for and against are discussed
in the following paragraphs in the context
of performance in the modern era since
2000 for both existing and new dams. It is
important to focus on the modern era since
much has changed and tailings manage-
ment is now widely characterised by much
improved governance and more thorough
engineering.
Arguments against upstreaming
The first of the arguments is that upstream-
ing is unsafe since the failure rate is higher
than for centreline and downstream raised
facilities. There have been several papers
which examine the statistics since 2000 and
do not show a definitive difference in failure
rate between upstream, centreline and
downstream raised facilities in the modern
era. The United Nations Environmental
Programme (UNEP) (Roche et al 2017)
provides statistics for the incidents that have
been recorded across the globe since 2000.
One of the interesting observations from
the UNEP study is that there have been no
significant incidents or failures in Southern
Africa (excluding Jagersfontein) where there
are more than 250 operational tailings
facilities of which 90% or more are raised by
upstreaming.
The second argument is that manage-
ment rigour is insufficient to sustain
consistent standards and that failures will
occur because of management failure.
This statement has merit since there has
been a loss of skill in the Southern African
mining industry and there are emerging
weaknesses in tailings operations. This
is an area where the Southern African
industry will need to strengthen capacity to
minimise risk.
The third argument is that if tailings
can liquefy, it must be assumed that it
will liquefy (Robertson 2021a), and that
upstream raising is therefore flawed since
the tailings of which the outer embank-
ment is formed will always have a pro-
pensity to liquefy. This is one of the most
controversial restrictions that has emerged
from the recent standards (Robertson
2021b) and will be one of the most difficult
to satisfy. Furthermore, if this require-
ment is applied strictly, many facilities
in Southern Africa will not satisfy the
safety criteria. However, dealing with this
Volume 65 Number 2 June 2023
requirement is not impossible for upstream
facilities, as we shall see later in the paper.
The fourth argument is that raising
with tailings does not meet with interna-
tional standards. Although this require-
ment does not appear in most standards, it
is implicit in the European Union guidance
(Cusano et al 2017). The justification for
the standard is, however, based on envi-
ronmental considerations and should not
preclude the Southern African practice of
using tailings for raising from continuing,
provided that the environmental conse-
quences are mitigated.
Arguments for upstreaming
The first argument for upstreaming is that
it has been practised relatively safely in
Southern Africa for more than 100 years.
The two failures that have occurred can
be attributed to excess water retention and
overtopping, and not for any of the reasons
cited by those against upstreaming.
The second argument is that Southern
African management practice has evolved
to suit labour and management intensive
requirements for upstreaming. In this, the
Southern African practices, reinforced sub-
stantially by SANS 10286 (SANS 1998), are,
in the author’s opinion, world-class, albeit at
a practice rather than governance level.
The third argument is that there are no
known cases of static or dynamic liquefac-
tion as the initiating cause of failure in
Southern Africa, and no failures of any
kind have been reported since SANS 10286
was published in 1998 (SANS 1998) (one
unreported case is known to the author). In
this period many significant failures have
occurred elsewhere in the world, including
significant failures in downstream and cen-
treline raised facilities. These failures have
also usually been followed by liquefaction
of the tailings.
The fourth argument is that the regions
where mining takes place in Southern
Africa have low seismic activity, being
located on a stable craton. The statisti-
cally probable peak ground acceleration
for a 1:10000 event (which is extrapolated
from a known database of seismic events)
is unlikely to exceed 0.15 g, although it
is acknowledged that this is currently
being challenged on theoretical grounds.
According to Vick (1983), based on case
histories, this level of acceleration is
unlikely to bring about liquefaction.
The fifth and last argument is that the
Southern African climate is favourable for
upstreaming. Use can be, and has been,
11
made of the natural climatic deficit to des-
iccate and stabilise the tailings to eliminate
liquefaction potential.
No compelling reason to ban
upstreaming in Southern Africa
The arguments for and against do not
provide a clear direction, but it is evident
that there is no compelling reason to ban
upstreaming in Southern Africa. There
are, however, conditions for proceeding.
Morgenstern (2018) refers to these condi-
tions as the key principles that must be
adhered to in the design, construction and
operation of tailings facilities. So, what
are these principles and how should we
be applying them? The following section
presents the principles to which Professor
Morgenstern was referring to, as ten rules.
TEN RULES FOR
UPSTREAM RAISING
The following rules for upstreaming have
been derived based on experience and
with reference to the published work of
Lenhart (1950), Vick (1992) and Martin and
McRoberts (1999). The rules have in some
instances been modified and summarised
by the author, and reference should there-
fore be made to the original publications
for the full descriptions.
1. A sufficiently wide beach, relative to
the ultimate height of the dam, must
be maintained to form a strong, wide,
drained (unsaturated), and/or dilatant
(non-contractive) outer shell (prism).
2. The prism upon which the stability
of the dam relies must not be com-
promised by underlying tailings that
are contractive or weak which, when
subjected to excess strain or seismic
loading, can either weaken or liquefy to
the point where the factor of safety is
insufficient.
3. The prism must be of sufficient width to
retain “bursting pressures” (Casagrande
& McIver 1970) of the contractive tail-
ings that may be located upstream of
the prism.
4. Where the tailings prism is relied on to
provide stability, the rate of raising of
the dam must be sufficiently slow, such
that there is a sufficient dissipation of
excess pore pressures and desaturation
in the supporting prism.
5. There must be sufficient underdrain-
age (drainage blanket, finger drains,
etc) and/or a pervious foundation to
maintain the prism in a drained and
12
Non-liquefiable
tailings prism
A D
Figure 1 Illustration of supporting prism
desaturated condition, and to prevent
seepage from issuing from the face of
the tailings dam.
6. Design analyses must include both
un­drained strength analysis and
effective stress analysis, with design
controlled by the analysis type that best
represents the anticipated behaviour
of the tailings. A wide range of factors,
including material type, degree of
consolidation, loading rate and stress
path, must be assessed in selecting the
appropriate analysis method.
7. A high degree of regular performance
monitoring, reviews and ongoing
involvement by the designer (or another
engineer formally appointed to assume
this role) is essential to confirm that
design intent is being satisfied.
8. Conventional upstream dams cannot
be considered for areas of moderate
seismicity without improved upstream
construction, involving a combina-
tion of compaction or other means of
densification (such as desiccation) of the
outer shell and good internal drainage.
Upstream dams should not be consid-
ered in areas with high seismicity.
9. The design must be consistent in terms
of design requirements (e.g. minimum
beach width) versus operational
requirements (e.g. pond size required
for clarification, storm storage and
freeboard). The geotechnical design of
upstream tailings dams should not be
carried out in ignorance of operating
constraints.
10. Seepage conditions within the dam
must be well-defined, requiring a good
understanding of hydraulic properties,
pore pressure profiles and hydraulic
gradients.
Although these rules have not been pre-
scribed by South African guidance and
regulations, they are generally understood
and followed. However, the author has
been exposed to a large proportion of the
operational Southern African tailings
facilities over the past 15 years – firstly,
as an operator with Fraser Alexander, and
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
B
Liquefiable
tailings
C
secondly, in the role of independent tailings
review board member. This experience has
shown that designers have begun to deviate
from the rules, often in minor ways, and
sometimes in significant and important
ways. For example, in the time before
designs were prepared by professional engi-
neers, facilities were limited to a maximum
height of 30 m based on empirical rules.
This limit made it relatively easy to achieve
a desaturated prism with width approxi-
mately equal to the final height. However,
as the height of facilities has increased well
above 30 m, the reason for the empirical
rule has been forgotten and the required
desaturated prism width has not been
maintained. This has been particularly
prevalent and concerning for facilities
raised with cyclones in the upstream mode.
There is a need therefore always to refer to
the ten rules, and to explicitly adopt them
in design.
The ten rules discussed
The outer shell or prism
The dimensions of the outer shell, which
must be maintained in a partially saturated
or dilatant state, need to be determined
by analysis. As a starting point, a support-
ing prism that forms a triangle extending
downstream from the final crest will be
adequate in most cases (illustrated as
triangle ABC in Figure 1). Optimisation
may reduce the required prism dimensions
to the triangle ABD, including reposition-
ing of B to prevent topping failures. The
extended prism has implications for drain
design which go beyond the current guid-
ance as offered in ICOLD Bulletin 97
(ICOLD 1994). This point also addresses
the second rule insofar as having a stable
prism will preclude the presence of under-
lying tailings that are contractive or weak.
Bursting pressure
The prism must be of sufficient width to
retain “bursting pressures” which can build
up behind upstream raises, as described
by Casagrande and McIver (1970) and

Downstream facing of dam:
Berms of soil or rockfill to
create outer wall of tailings
dam, placed on top of
one another as the dam
increases in height
The arrows represent bursting
pressures acting on the virtual plane
Figure 2 Illustration of bursting pressures (adapted from Casagrande & McIver (1970))
illustrated in Figure 2. This mechanism is
seldom evaluated but needs specific atten-
tion, especially for existing upstream raises
which are formed from earth or other
materials against which hydrostatic pres-
sure can develop. Casagrande and McIver
(1970) go on to say that “… the practice of
raising tailings dams by a series of dykes,
each supported in part by the underlying
dyke and in part on soft or loose tailings,
does not assure adequate safety. This
construction practice can provide safe dams
only when the sand fractions of the efflu-
ent … are separated and compacted to at
least the axis of the final dike of the dam.”
The axis of the final dike referred to by
Casagrande and McIver (1970) is illustrated
by line BC in Figure 1.
Rate of rise
Rate of rise, tailings layer thickness and
cycle management are determinants of the
rate of desiccation of tailings in a particu-
lar climatic setting. The rate of rise and
the layer thickness and cycle time must
be established to achieve the required
degree of desaturation to preclude lique-
faction. The degree of desaturation that is
required to ensure that contraction during
shearing does not result in pore pressure
build up when subjected to straining or
seismic loading is not well established.
Hence a maximum degree of satura-
tion of 85% is usually assumed, unless a
higher percentage can be demonstrated to
suffice.
Internal drainage
Sufficient underdrainage must be provided
to maintain the prism in a drained state.
The facility geometry and location of the
Journal of the South African Institution of Civil Engineering
Virtual (imaginary) plane on Water level on top of tailings
which bursting pressure acts
γw+t H be designed to measure performance for
Ground on which dam is built
Tailings
pond play an important role in determining
the extent and arrangement of drainage.
Drainage needs to be more extensive to
bring about drawdown where the pool
is located close to the crest, and where
layering, which leads to higher horizontal
than vertical permeability, may require
the use of chimney or curtain drains that
interrupt the continuity of the horizontal
layers to intercept seepage moving laterally.
To provide for the required drainage, seep-
age conditions must be well defined and
understood. This requires comprehensive
analysis by specialised personnel.
Analysis
The analyses must include both undrained
and drained analysis as a starting point.
A drained analysis provides a measure
of the margin of safety against conven-
tional loading conditions in the absence of
any events that might trigger undrained
behaviour. It remains the most reliable
method to establish the margin of safety.
An undrained analysis, on the other
hand, indicates the margin of safety against
abnormal loading conditions causing a
failure. The most conservative approach
is to assume that the worst-case loading
conditions can arise, and to design accord-
ingly. In many cases with existing facilities,
it is not practicable to assume the worst
case, and it becomes necessary to examine
the probability of occurrence of abnormal
loading conditions. Seismic loading is one
of the loading conditions that can induce
undrained behaviour, and hence, where
the tailings can liquefy, upstreaming
should not be considered in seismically
active areas without improved upstream
construction.
Volume 65 Number 2 June 2023
Performance monitoring
It goes without saying that a high degree of
regular and rigorous performance monitor-
ing is required for upstream raised facilities
to provide assurance that the development
of the facility is in conformance with the
design requirements. Monitoring should
all possible failure mechanisms and to
provide early warning of developments that
indicate a trend towards non-conformance.
Monitoring cannot provide early warning
of undrained failure and therefore the
monitoring system should be designed to
measure the parameters that must change
to initiate an undrained failure. Monitoring
of earthquakes is, however, not useful as an
observational tool.
Consistency between
design requirements and
operational outcomes
The design requirements must be explicitly
stated by the designers so that performance
indicators can be set and monitored dur-
ing operations. For example, a minimum
beach width is assumed for the drainage
design and stability analysis but is often
only shown on cross-sections that appear
in the appendices of design reports. The
minimum dry beach width therefore
does not get to be established as a formal
performance requirement. The process of
crystalising design assumptions into opera-
tional performance criteria requires a good
understanding of the design models and
of the upset conditions that could result
in non-conformance. Senior oversight of
design is particularly important for identi-
fying these requirements before operations
commence.
MEETING THE REQUIREMENTS
IN SOUTHERN AFRICA
The ten rules can be met for new upstream
designs without too much additional cost.
It is therefore strongly advocated that
designers should fully internalise the rules
and ensure that they are built into future
designs. Independent reviewers should also
focus on the rules and should be specifical-
ly requested to provide opinion on whether
the rules have been incorporated into the
design and, if not, what the risks are.
Many of the existing facilities in
Southern Africa will not meet the require-
ments, principally because they will fail on
rules 1 and 2. The prisms that have been
designed to support the weak tailings in
13
the interior are often undersized owing to
incorrect positioning of the drains, failure
Urgent action including
of the drains or inadequate gradients. This
QQ Low FOS and
Drained consideration of
QQ High probability of failure
then leads to a situation where the critical cessation of operations
failure surface passes through tailings that
QQ Low FOS and Urgent remedy including
have inadequate peak and/or post-peak Undrained QQ High probability of failure consideration of
undrained shear strength and where the peak
for credible triggers cessation of operations
factor of safety criteria are not met.
The real risk is, however, somewhat High probability of liquefaction
Residual or
and loss of containment Timely remedy
different. Several tailings facilities exist in liquefied
due to extreme event
Southern Africa with undrained factors
of safety of less than 1 for peak undrained
strength with an implied probability of No action

failure of close to certainty but have never

failed even after exposure to the largest
mining induced tremors of the Free State
and Witwatersrand. This fact clearly calls
into question the approach that is cur-
rently adopted globally to deal with the
undrained condition. The shortcoming,
in the author’s opinion, is associated with
a fixation on “compliance” or rule-based
approach as opposed to a risk-based
approach. A risk-based approach would
recognise that a low factor of safety for
an undrained condition is a flag which
prompts the designer to examine the trig-
gers that could bring about an undrained
condition, and to assess the potential
for these conditions to develop. Thus, in
moderate and high seismic regions the
designer would most likely conclude that
the probability of occurrence of an earth-
quake of sufficient magnitude that would
lead to failure is too high, while elsewhere
the designer might conclude that the
probability of occurrence of triggers for
undrained behaviour, such as removal of
support (associated with a conventional
failure) or a rapid increase in phreatic sur-
face that could trigger an undrained fail-
ure, are just too low to warrant concern.
Unfortunately, the rule-based approach is
prevalent and is favoured by many non-
Southern African reviewers. This requires
a concerted effort by Southern African
engineers to demonstrate that risks
associated with upstreaming are within
reasonable limits.
The Global Industry Standard on
Tailings Management (GTR 2020) makes
provision in Requirement 4.7 for resolution
of the impasse when target factors of safety
are not met. Requirement 4.7 states that
existing tailings facilities shall conform
to the minimum requirements, except for
those aspects where the Engineer of Record
(EoR), with review by the Independent
Tailings Review Board or a senior inde-
pendent technical reviewer, determines
Figure 3 Hierarchical process for stability evaluation
that the upgrade of an existing tailings
facility is not viable or cannot be retroac-
tively applied. In this case, the Accountable
Executive shall approve and document
the implementation of measures to reduce
both the probability and the consequences
of a tailings facility failure to reduce the
risk to a level as low as reasonably practi-
cable (ALARP). The basis and timing for
addressing the upgrade of existing tailings
facilities shall be risk-informed and carried
out as soon as reasonably practicable.
Most Engineers of Record are having
difficulty in implementing Requirement
4.7 since there is not an established process
to do this and very little thought has been
given as to how to assess the probability of
occurrence of most of the trigger mecha-
nisms. The process should, however, not
be complicated and should follow the
hierarchical approach which is illustrated in
Figure 3 to facilitate stability evaluation and
interpret factors of safety (FoS) in context.
The process follows a series of steps
beginning with confirmation that either
the factor of safety exceeds the target
threshold or that probability of failure for
conventional drained stability is low (say
<1:1000 to 1:10000). This would represent
the lower bound for engineering reliability
and may be taken to demonstrate that, but
for unforeseen circumstances, the facility
safety is adequate. If the target factor of
safety is not adequate, then the probability
of failure should be determined to decide
whether mitigation is required to increase
the factor of safety.
If the drained factor of safety or
the probability of failure is found to be
adequate, then proceed to the next step to
determine the undrained factor of safety.
As for the first step, if the factor of safety is
not adequate, confirm that the probability
of triggers exceeding a level that would
induce undrained behaviour (say again
<1:1000 to 1:10000) would be acceptable
and, if not, develop and implement reme-
dial measures.
If the undrained factor of safety is
adequate or the probability of inducing
undrained behaviour is low enough, then
proceed to the last step. If the tailings can-
not liquefy or if the probability of liquefac-
tion is low enough, then no further action
is required. If the tailings can liquefy, then
the risk should be mitigated by an ALARP
process. The term “can liquefy” should be
defined as being possible with a probability
of >1:1000 to 1:10000.
Actions which can be considered in
the ALARP process include increasing the
factor of safety to >1.1 for post-earthquake
shear strength by buttressing, strength-
ening or slope flattening; providing for
protection of people by early warning;
physical containment and or diversion
infrastructure; or, as a last resort, resettling
people and relocating infrastructure in the
inundation zone.
PRACTICAL APPLICATION
OF THE ALARP PROCESS
GISTM Requirement 4.7 stipulates that the
Accountable Executive shall approve and
document the implementation of measures
to reduce both the probability and the
consequences of a tailings facility failure to
reduce the risk to a level as low as reasonably
practicable (ALARP). The basis and timing
for addressing the upgrade of existing tail-
ings facilities shall be risk-informed and car-
ried out as soon as reasonably practicable.
ALARP requires that all reasonable
measures be taken with respect to ‘toler-
able’ or ‘acceptable’ risks to reduce them

14 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering

Failure modes
Level 1
Screen out
Example of
process to
Carry
determine
ALARP
Level 2
Mitigate
Assess
Figure 4 Simplified ALARP process
even further until the cost and other
impacts of additional risk reduction are
grossly disproportionate to the benefit.
The ALARP process must be carried
out in a systematic manner and should be
documented to provide an audit trail and
serve as the basis for justification to regula-
tors and stakeholders that the process has
been thorough, and that the outcome is
defensible. The principles defining the
process are not yet settled and there are
many ways in which an outcome could be
achieved. Wates (2021) proposed a simpli-
fied process that is illustrated in Figure 4.
In the context of ALARP, tolerable risk,
as defined in ICOLD Bulletin 130 (ICOLD
2005) and adapted from the United
Kingdom Health and Safety Executive
definition, is “… a risk within a range that
society can live with to secure certain net
benefits. It is a range of risk that we do not
regard as negligible or as something we
might ignore, but rather as something we
need to keep under review and reduce it still
further if we can.”
In any assessment as to whether risks
have been reduced to ALARP, measures
to reduce risk can be ruled out only if the
sacrifice involved in taking them would be
grossly disproportionate to the benefits of
the risk reduction.
CONCLUSION
The call to eliminate or ban upstreaming is
not justified in regions with low seismicity.
Provided that the ten rules for upstreaming
Journal of the South African Institution of Civil Engineering
Define failure modes and consequences
Prepare Level 1 risk assessment (qualitative)
Screen out the non-credible and lesser risks for which the
assessment has an acceptable degree of reliability
Carry risks with a high degree of uncertainty (i.e. they could fall
into high-risk category) and high risks into Level 2 assessment
Prepare Level 2 risk assessment (semi-quantitative or
quantitative)
Develop mitigation measures for high risks
Prepare costs and benefit for high risk and propose viable
mitigation to achieve ALARP
are applied, there is no reason why future
upstream raised facilities cannot be
designed and constructed to be as safe as
centreline and downstream facilities.
Eliminating upstreaming in Southern
Africa is not justified by the record of
failures of upstream facilities in the region.
The cost of shutting down and replacing
existing facilities associated with operating
mines would be ruinous for most mines,
while the risk has been shown by the
record to be reasonably low.
The continued use of existing upstream
raised facilities can be motivated, provided
there is strong engineering backing and
comprehensive assessment of the probabil-
ity of the development of abnormal trigger
mechanisms that could induce undrained
behaviour. The motivation for continued
use of facilities that do not comply with
the minimum factors of safety, but with
demonstrated acceptable margins of safety,
must be underpinned by a comprehensive
ALARP process that provides a defensi-
ble argument for the mitigation that is
implemented.
Obsession with the assumption that
static liquefaction is the most significant
risk for upstream tailings facilities intro-
duces additional risk, since this distraction
could lead to the key mechanisms being
overlooked. It is critical to point out that a
design can be flawed (perhaps even fatally)
if important, plausible failure mechanisms
are not considered and that factors of
safety cannot fully account for gross omis-
sions in design and analysis. Designers
Volume 65 Number 2 June 2023
and Engineers of Record should not be
distracted by focusing blindly on meeting
the standards.
ACKNOWLEDGEMENT
This paper is based on a presentation deliv-
ered by the author to the SAIMM Global
Tailings Standards and Opportunities for
the Mine of the Future online conference
(15, 17, 19, 22, 24, 26 November 2021) as
the recipient of the SAICE Tailings Medal
for 2021.
REFERENCES
Boswell, J & Sobkowicz, J 2018. Leading versus lagging
indicators of tailings dam integrity. Proceedings,
22nd International Conference on Tailings and
Mine Waste. 30 September – 2 October 2018,
Keystone, CO.
Casagrande, L & McIver, B N 1970. Design and
construction of tailings dams. In Brawner, C O &
Milligan, V (Eds). Proceedings, 1st International
Conference on Stability in Open Pit Mining,
American Institute of Mining, Metallurgical
and Petroleum Engineers. New York, Vol 11,
pp 181–203.
Cusano, G, Gonzalo, M R, Farrell, F & Remus, R
2017. Best Available Techniques (BAT) reference
document for the non-ferrous metals industries.
Brussels: EIPPCB (European Integrated Pollution
Prevention and Control Bureau). Available at:
https://eippcb.jrc.ec.europa.eu/reference/
non-ferrous-metals-industries-0.
GTR (Global Tailings Review) 2020. Global
industry standard on tailings management.
Available at: https://globaltailingsreview.org/
global-industry-standard.
ICOLD (International Commission on Large Dams)
1994. Tailings dams design of drainage. Review and
recommendations. Bulletin 97. Paris: ICOLD.
ICOLD 2005. Risk assessment in dam safety
management. A reconnaissance of benefits. methods
and current applications. Bulletin 130. Paris:
ICOLD.
Jennings, J E 1979. Mechanisms of failure and
associated design considerations: The failure of a
slimes dam at Bafokeng. The Civil Engineer in South
Africa, 21(16): 135–145.
Lenhart, W 1950. Control of tailings from washing
plants. Rock Products, 53(7): 72–80.
Martin, T E & McRoberts, E C 1999. Some
considerations in the stability analysis of upstream
tailings dams. Proceedings, 13th International
Conference on Tailings and Mine Waste. Banff,
Canada, January 1999.
McRoberts, E C T, MacGowan, R, Eenkooren, R &
Pollock, G 2017. 50 years of successful learnings
by experience: Suncor tailings geotechnical.
15
 Proceedings, 21st International Conference
on Tailings and Mine Waste. Banff, Canada,
pp 229–239.
Morgenstern, N R 2018. Geotechnical risk, regulation,
and public policy. Victor de Mello Lecture. Soils
and Rocks, 41(2): 107–129.
Roche, C, Thygesen, K & Baker, E (Eds) 2017. Mine
tailings storage: Safety is no accident. A UNEP
Rapid Response Assessment. United Nations
Environment Programme and GRID-Arendal,
Nairobi and Arendal. Available at: http://www.
grida.no.
16
Robertson, P K 2021a. Evaluation of flow liquefaction
and liquefied strength using the cone penetration
test: An update. Canadian Geotechnical Journal,
59(4). doi: 10.1139/cgj-2020-0657.
Robertson, P K 2021b. Can you safely construct a
tailings facility using upstream construction. 18th
Jennings Lecture. SAICE Geotechnical Division.
Available at: https://www.youtube.com/channel/
UCZelzYmgXIXSsed_pXWvEUg.
SANS (South African National Standard) 1998. SANS
10286: Mine Residue. Pretoria: SABS Standards
Division.
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
Vick, S G 1983. Planning, Design, and Analysis of
Tailings Dams. Wiley Series in Geotechnical
Engineering. New York: Wiley.
Vick, S G 1992. Stability evaluation during staged
construction: A discussion. Journal of Geotechnical
Engineering, 118: 1282–1289.
Wagener, F 1997. The Merriespruit slimes dam failure.
Overview and lessons learnt. Journal of the South
African Institution of Civil Engineering, 39(3): 11–15.
Wates, J A 2021. Practical application of ALARP.
Presentation to ICOLD Meeting. Santiago, Chile.
Available at: https://www.researchgate.net.
Could a one-size-fits‑all TECHNICAL PAPER
approach apply to the JOURNAL OF THE SOUTH AFRICAN
INSTITUTION OF CIVIL ENGINEERING
extension of stage‑discharge ISSN 1021-2019 (print) | ISSN 2309-8775 (online)
Vol 65 No 2, June 2023, Pages 17–27, Paper 1568

relationships at PROF JACO GERICKE (Pr Eng, Int PE (SA)), is Head

flow‑gauging weirs?
of the Department of Civil Engineering at the
Central University of Technology, Free State
(CUT). He obtained his PhD Eng (Agriculture)
degree from the University of KwaZulu-Natal in
2016 and has more than 25 years’ professional
and academic experience, mainly in the fields of
O J Gericke, V H Williams
hydrology, water resources management, and
river hydraulics. He has published several papers in the design hydrology
field and is involved in research projects funded by the Water Research
At a flow-gauging weir, the stage or flow depth is normally measured continuously and Commission.

converted into discharge using a stage-discharge (SD) rating curve (RC). During flood Contact details:
events, the observed water levels often exceed the flow-gauging weir’s designed measuring Department of Civil Engineering
Faculty of Engineering, Built Environment and Information Technology
capacity or structural limit. Subsequently, the standard calibration of the flow-gauging weir
Central University of Technology, Free State
becomes irrelevant and the extension of the SD RC for above-structure-limit flow conditions Bloemfontein
is required. This paper attempts to identify a one-size-fits-all approach for the extension of South Africa
SD RCs by assessing seven indirect hydraulic extension methods and a one-dimensional HEC- T: +27 51 507 3516
E: jgericke@cut.ac.za
RAS modelling approach against direct SD measurements or extension methods at selected
flow-gauging sites in South Africa. In considering a ranking-based selection procedure and
VERNON WILLIAMS is a civil engineering
associated goodness-of-fit (GOF) criteria, the stepped backwater analysis, slope-area method, technologist and lecturer currently employed by
and 1-D HEC-RAS steady flow modelling proved to be the most appropriate. The other indirect the Tshwane University of Technology (TUT). He
extension methods resulted in larger statistical differences between the at-site benchmark and graduated with BTech Eng (Civil) and MEng
(Civil) degrees from the Central University of
modelled values. Given that the extension of RCs is significantly more affected by the site and
Technology, Free State (CUT), and has a particular
river reach geometry, initial hydraulic conditions, flow regimes, and level of submergence at interest in flood hydrology and river hydraulics.
high discharges than the actual extension method used, it was confirmed that a one-size-fits-all He currently has six years’ academic experience.
approach cannot be applied for the extension of SD RCs in South Africa. Contact details:
Department of Civil Engineering
Faculty of Engineering, Built Environment and Information Technology
Central University of Technology, Free State
INTRODUCTION limit (Petersen-Øverleir & Reitan 2009). Bloemfontein
The first flow-gauging weirs were con- Hence, the standard calibration of the South Africa
structed in 1904 in the Transvaal Province flow-gauging weir will no longer be a true T: +27 12 382 2458
E: williamsvh@tut.ac.za
to measure streamflow continuously in reflection of the actual discharges that
South Africa (Menné 1960). By the year had occurred during the flood events. The
2007, streamflow data had already been latter observed stage levels exceeding the
recorded continuously at 782 different maximum rated stage or flow depth (H)
Department of Water and Sanitation is a common problem in South Africa, as
(DWS) flow-gauging sites, comprising shown in Figure 1, where the RC exceed-
sharp-crested weirs (55%), crump weirs ances and missing data are expressed as
(35%), and the remaining 10% consisting a percentage of the annual maximum
of broad-crested weirs, dam spillways, series (AMS) at 806 DWS flow-gauging
and velocity-area gauging sites (Wessels & weirs with 20 or more years of record at
Rooseboom 2009). However, streamflow each. For example, a standard SD RC, as
is seldom directly measured; instead, the shown in Figure 2, must then be extended
stage or flow depth is measured at regular beyond H to reflect the high discharges (Q)
intervals and converted into discharge exceeding the designed measuring capacity.
using a stage-discharge (SD) rating curve In general, such extensions are normally
(RC) at a flow-gauging weir or specific limited to bankfull flow conditions using
river section. appropriate indirect methods, while high
During flood events, the observed water flow extensions above bankfull flow condi-
levels often exceed the flow-gauging weir’s tions should only be considered in cases
designed measuring capacity or structural where the existing SD relationship includes

Keywords: above-structure-limit, extension, flow-gauging weir,

Gericke OJ, Williams VH. Could a one-size-fits-all approach apply to the extension of stage-discharge relationships at flow-gauging weirs? rating curve, stage-discharge relationship
J. S. Afr. Inst. Civ. Eng. 2023:65(2), Art. #1568, 11 pages. http://dx.doi.org/10.17159/2309-8775/2023/v65n2a3

The Journal of the South African Institution of Civil Engineering, which is distributed internationally, is a peer-reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND). Readers may therefore freely use and share the content as long as they credit the original creator and publisher, do
not change the material in any way, and do not use it commercially. Copyright of this article remains with the authors. 17
floodplain flow on the full width of the
floodplain (Gericke & Smithers 2017). 35
Direct measurements, e.g. conventional
current gaugings, are not always possible 30 29.7
owing to various practical constraints asso-
ciated with high discharge events, e.g. high 25 24.6

Flow-gauging weirs (%)

velocities and water depths, danger to staff 21.9 21.3
and equipment entering a required river
20
reach, and operational difficulties (Lang 17.4 17.1 16.4
et al 2010). Therefore, developing reliable
SD relationships at high discharges using 15
13.1 13.0
direct measuring techniques becomes
problematic (Lindner & Miller 2012). As 10 8.9
a result, different indirect methods for 6.3 5.8
extending SD RCs are available; however, 5 4.1
the impact of using these different methods
varies significantly and highlights the need 0.5
0
for a robust and reliable extension method, 0 0–10 11–20 21–30 31–40 41–50 51–100
since significant errors, uncertainty, and AMS values (%)
variable systematic discharge bias could be Percentage RC exceedance Percentage missing data
introduced (Lang et al 2010). For example,
in flood hydrology, where probabilistic Figure 1 R
 C exceedances and missing data at 806 DWS flow-gauging stations with record lengths
flood frequency analyses using the AMS > 20 years (after Nathanael 2016; Smithers 2022)
are conducted to estimate design peak dis-
charges, the exclusion of above-structure-
limit discharges typically would result 250
in the underestimation of flood events,
while the inappropriate extension thereof
could introduce uncertainty and ordered 200
bias (Gericke & Smithers 2017). The latter
Discharge (Q, m3/s)

is quite evident in AMS data sets having

different plotting positions for flood events 150

of equal magnitude, i.e. cases where the

above-structure-limit of the flow-gauging
weir was exceeded, and the standard RC (as
shown in Figure 2) becomes irrelevant.
Various international studies were
undertaken in an attempt to develop
robust and reliable RC extension meth-
ods. For example, the Environmental
Agency in the United Kingdom (EA UK),
undertook a study entitled Extension of
Rating Curves at Flow-gauging Stations
to compile a best practices manual using
hydraulic and computational modelling
techniques (Ramsbottom & Whitlow 2003).
Dymond and Christian (1982) evaluated
the accuracy of SD RCs by considering
both the individual and average discharge
measurement errors during a specific
period. Petersen-Øverleir and Reitan (2009)
examined the joint impact of sample
variability and RC inaccuracy on at-site
flood frequency analysis. Lang et al (2010)
extended SD RCs using hydraulic model-
ling to ultimately improve flood frequency
analyses. Shao et al (2018) extended SD
RCs using hydrodynamic models and, as a
result, quantified the uncertainty associ-
ated with the overall process.
100
50
0
0 0.5
Figure 2 E xample of a standard RC with a structural limit of H = 3.4 m and Q = 225.3 m3/s at U2H057
(https://www.dws.gov.za/Hydrology/Verified/)
However, despite all the above studies,
there are currently no user manual or stan-
dard practices available to extend SD RC
relationships in South Africa. The absence
of standard practices to extend SD RCs
is ascribed to the different and unique
hydraulic characteristics, topography, and
discharge conditions present at South
African flow-gauging sites; hence, site
visits and a fundamental understanding of
the at-site hydraulics are crucial to ensure
the accurate extension of SD RCs. In the
absence of sufficient (high resolution) topo-
graphical data and site surveys, Gericke
1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Stage (H, m)
and Smithers (2017) recommended that
individual stage extensions (HE), whether
for bankfull or above bankfull flow condi-
tions, should be limited to a maximum of
20%, i.e. HE ≤ 1.2 H. In the case of above
bankfull flow conditions, the relevance of
the general extension procedure described
above should be tested and compared
to other relevant hydraulic extension
methods, e.g. stepped backwater analysis
(SBA) and/or slope-area method (SAM) if
surveyed cross-section data is available.
In addition to the above-mentioned 20%
limit, the hydrograph shape, especially the

18 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
peakedness due to a steep rising limb in
relation to the hydrograph base length, and
the relationship between the observed peak
discharge (QPxi) and associated direct run-
off (QDxi) values, should be used as addi-
tional criteria to justify the HE extensions
up to 20%. Typically, in such cases, the
additional volume of direct runoff (QDE)
due to the extension should be limited to
5%, i.e. QDE ≤ 0.05 QDxi.
Given that the use of appropriate SD RC
extension methods is warranted not only
to enhance the estimation of design flood
events, i.e. peak discharges, but also to
impact on water resources management,
this paper attempts to identify a one‑size-
fits-all approach for the extension of
SD RCs by assessing a selection of indirect
extension methods (e.g. hydraulic and one-
dimensional modelling approaches) against
direct extension (benchmark) methods
(e.g. at-site conventional current gaugings,
hydrograph analyses, and level pool rout-
ing techniques) at selected flow-gauging
sites in South Africa. In this paper, it is
assumed that the SAM, and by extension
the divided channel method (DCM), are
more appropriate and hydraulically cor-
rect than the other SD extension methods
available. However, the SBA is regarded
as being more accurate than the SAM,
while one-dimensional hydraulic models
(e.g. HEC-RAS) better illustrate the actual
hydraulic conditions in a river reach than
simple hydraulic methods.
MATERIALS AND METHODS
A review of the seven indirect hydraulic
extension methods and one-dimensional
HEC-RAS modelling is discussed in this
section, followed by an overview of the
location and characteristics of the selected
flow-gauging sites. Thereafter, the method-
ologies adopted are expanded on in detail,
followed by the results, discussion and
conclusions.
Review of indirect hydraulic
extension methods
It is essential that a hydrometric data,
gauging, and rating review be conducted
prior to extending any SD relationship
(Ramsbottom & Whitlow 2003). The fol-
lowing hydraulic methods were considered
and applied at each site:
a. Simple extension (SE): The SE uses
the rating equation for the highest
limb of the RC to extend the SD
relationship to stages exceeding the
Journal of the South African Institution of Civil Engineering
above-structure-limit, or conversely, for
the lowest segment in the case of lower
stages (EA UK 2002). The application
of the SE is limited to uniform channel
cross-sections and bankfull flow condi-
tions throughout the entire RC range
(EA UK 2003). The SE is unsuitable
for: (i) above bankfull flow conditions,
and (ii) any transitions from in-channel
flow to overbank full flow conditions,
modular to submerged flow conditions
and vice versa, submerged flow to
complete channel control flow, or where
any changes in vegetation and cross-
sections occur.
b. Logarithmic extension (LE): The LE
has similarities to the SE; however,
the stage and discharge are plotted
on a logarithmic scale. The advantage
of the LE is that it can accommodate
changes in the slope of the rising limb
of an RC (EA UK 2002; Ramsbottom &
Whitlow 2003). According to the British
Standards Institution International
Organization for Standardization (BSI
ISO 1998), the LE is not recommended,
given that it can generate substantial
errors and is regarded as inferior to the
SE in many ways. Essentially, the LE is
a modified version of the SE limited to
a first-order polynomial function. The
same hydraulic conditions, limitations,
and data requirements as applicable to
the SE, apply to the LE.
c. Velocity extension (VE) methods: The
VE methods can only accommodate
limited changes in the channel section
geometry for above-structure-limit con-
ditions, while transitions from modular
to submerged flow conditions are nor-
mally poorly presented (EA UK 2003).
The hydraulic conditions, which are
regarded as unsuitable for the SE, also
apply to the following VE methods:
i. Velocity extension simple approach
(VE-SA): The VE-SA requires the
computation of the cross-sectional
area and velocity at incremental river
stages similar to the stages used to
derive the existing SD relationship.
The RC extension is based on an
extension of velocity against stage.
One cross-section is required at the
flow-gauging site to calculate the
change in cross-sectional area with
stage. Once the velocity (v) and cross-
sectional area (A) have been derived,
the discharge (Q) can be determined
using the conservation of mass
(Ramsbottom & Whitlow 2003).
Volume 65 Number 2 June 2023
ii.  Velocity extension hydraulic radius
approach (VE-HRA): In using the
VE-HRA, a plot of velocity against
hydraulic radius (R) frequently
presents a linear relationship; this
can also be used to provide values of
velocity in the extended range of the
RC. Once v and A have been derived,
Q can be determined using the
conservation of mass (Ramsbottom
& Whitlow 2003).
iii. Velocity extension Manning’s
approach (VE-MA): In using the
VE-MA, the Manning’s equation is
used by assuming that the square
root of the channel slope (So1/2) and
Manning’s roughness coefficient
(n) are constant across the full
range of discharges when compared
to AR2/3. The relationship of Q
versus AR2/3 is then extended to
provide the SD relationships for
above-structure-limit conditions
(Ramsbottom & Whitlow 2003).
d. Slope-area method (SAM): The
SAM considers the hydraulic radius,
cross-­sectional area, channel bed
(water surface) slope, and roughness
coefficients (e.g. Manning’s n-values
or Chézy’s ks-values) to determine the
SD relationship at a given site. Two
approaches are associated with SAM, i.e.
the simple approach (SA) and the DCM
(EA UK 2002). In using the SAM-SA,
the cross-sectional areas and water
surface slopes are determined for a range
of discharges, while the mean velocity
and discharge in the channel are calcu-
lated using Manning’s or Chézy’s open-­
channel flow equation (EA UK 2002).
The SAM-DCM requires surveyed cross-
sections that include the main channel
and floodplains. The surveyed cross-
sections are divided into three panels
using vertical division lines, i.e. the left
bank floodplain, main channel, and the
right bank floodplain. The channel rat-
ing is extended above bankfull, assuming
no friction on the vertical boundaries
for above bankfull flow conditions.
Separate ratings are calculated for each
component using different roughness
coefficient values and summed to obtain
a rating for the section (Ramsbottom &
Whitlow 2003). Although being regarded
as a much more versatile method and
less sensitive than the SE, LE, and VE
methods, similar hydraulic conditions,
limitations, and data requirements could
apply to the SAM.
19
Figure 3 Location of the 10 flow-gauging sites in South Africa (Williams 2023)
e. Stepped backwater analysis (SBA):
The SBA is based on the standard
step method and applies to the
gradually varied flow (GVF) regime in
non-prismatic channels. Field survey
data is required, e.g. the longitudinal
river profile, cross-sections, and a
detailed description of the roughness
coefficients applicable to each cross-
section (Chow 1959). The SBA calcula-
tions are done on a trial-and-error
(iterative) basis, which can be easily
automated using solver functions in
Microsoft Excel.
f. 1-D HEC-RAS modelling: The United
States Army Corps of Engineers
(USACE) developed the Hydraulic
Engineering Centre River Analysis
System (HEC-RAS) software, with
Version 1.0 having been released in
August 1995 (Maeder 2015). The
software can perform one-­dimensional
steady flow, two-dimensional unsteady
flow, sediment transport, and water
temperature or quality modelling. The
steady flow analysis in HEC-RAS is
based on the GVF theory. In applying
the energy equation, energy losses are
20
assessed by friction and contraction or
expansion coefficients. The momentum
equation applies to situations where
the water surface profile varies rapidly,
e.g. hydraulic jumps, flow conditions
at bridges, and river confluences
(USACE 2016).
Flow-gauging sites
Initially, all the SD extension reports avail-
able from DWS, i.e. the Discharge Table
Improvement Reports, were studied, given
that a standard SD rating table is regarded
as the first step towards the extension
of any SD relationship. Subsequently, a
total of 10 flow-gauging sites, as shown in
Figure 3, were identified and selected.
In terms of the hydrometric and geo-
metric data requirements, the selection of
the flow-gauging sites took the following
aspects into consideration:
a. Streamflow data: Record length,
data quality, flow duration curves to
highlight the occurrence and frequency
of minimum and maximum flow
ranges, and the number of standard
and extended SD relationship tables or
curves available.
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
b. Hydraulic conditions: Modular and/
or submerged flow conditions, variable
submergence due to backwater effects
and vegetative growth, identification of
possible hydraulic controls, sediment
transport, unsteady flow conditions,
and the influence of in-bank and out-of-
bank flow paths.
c. Geometric properties: Type of flow-
gauging weir, e.g. sharp or broad-crest-
ed, crump, hydro-flumes, broad-crested
flank walls, flood or natural sections,
overall river topography and layout,
river channel and flood plain geometry,
position of control points within the
river system (especially at high flows),
availability of survey data (e.g. cross-
sections, longitudinal sections, previous
flood surveys, and the previous or cur-
rent allocation of site-specific roughness
coefficients).
Typically, the chosen flow-gauging sites in
Figure 3 are characterised by complex flow
(hydraulic) conditions and are not neces-
sarily sites that had gradual changes in
geometric properties. Ultimately, the sites
chosen are inclusive of the following prob-
lem areas: (i) submergence, (ii) estimation
Table 1 Summary of flow-gauging site characteristics (Williams 2023)
Latitude Period River Type Structural Benchmark Indirect
Site
Longitude of record Province of structure limit (Q & H) method methods
Compound weir
24°04’58”S 1962/08/27 Mokolo Q = 115 m3/s Level pool
A4H005 5 Sharp crests SE, LE & VE
27°46’23”E to present Limpopo H = 1.4 m (back) routing
6 Broad crests
22°32’07”S 1995/02/02 Mogalakwena Compound weir Q = 153 m3/s Level pool
A6H035 SE, LE & VE
28°53’51”E to present Limpopo 4 Sharp crests H = 1.3 m (back) routing
C2H003
C2H003 28°49’11”S 1924/01/29 Vaal Compound weir Q = 481 m3/s Level pool
SE, LE & VE
(C2H130) 28°03’49”E to present Gauteng 8 Sharp crests H = 2.2 m (back) routing
C2H130 flood section (FS)
Compound weir SE, LE, VE,
29°02’33”S 1969/07/03 Riet Q = 90 m3/s
C5H014 Broad-crest SBA SAM, SBA &
24°35’57”E to present Free State H=1m
V-crump HEC-RAS
Compound weir SE, LE, VE,
33°53’52”S 1970/03/24 Bree Q = 81 m3/s SBA
H5H004 Hydro-flume SAM, SBA &
20°00’46”E to present Western Cape H = 1.4 m
2 Sharp crests HEC-RAS
Compound weir SE, LE, VE,
33°41’05”S 1982/04/07 Touws Q = 131 m3/s
J1H018 V-crump (low) SBA SAM, SBA &
21°08’46”E to present Western Cape H=2m
4 Crumps HEC-RAS
SE, LE, VE,
34°01’02”S 1964/05/01 Gourits Natural flood section (FS)
J4H002 N/A SBA SAM, SBA &
21°45’24”E to present Western Cape at bridge
HEC-RAS
34°01’40”S 1961/05/05 Groot Brak Compound weir Q = 55 m3/s Wolwedans
K2H002 SE, LE & VE
22°13’21”E to present Western Cape 4 Sharp crests H = 1.5 m Dam spillage
Compound weir
28°43’15”S 1967/08/01 Tugela Q = 98 m3/s Hydrograph
V1H026 4 Sharp crests SE, LE & VE
29°21’33”E to present KwaZulu-Natal H = 1.2 m analysis
Hydro-flume
Compound weir
SE, LE, VE,
24°46’08”S 1967/09/01 Sand Sharp crest Q = 76 m3/s
X3H008 SBA SAM, SBA &
31°23’24”E to present Mpumalanga Hydro-flume H = 1.95 m
HEC-RAS
Broad-crested flank walls

problems, (iii) complex geometry, and
(iv) unsuitable sites (e.g. J4H002) for the
construction of flow-gauging weirs. Given
the complex conditions and lack of suf-
ficient geometric data available at some of
these flow-gauging sites, not all the above-
listed indirect hydraulic extension methods
could be applied. Table 1 contains a sum-
mary of all the relevant characteristics
present at each flow-gauging site.
Processing of geometric data
The processing of geometric data, e.g. wet-
ted perimeter, wetted area, and hydraulic
radius, was done using the Windows Cross-
Section Professional (WinXSPRO) (Hardy
et al 2005), which is essentially a channel
cross-section analyser. Prior to the use of
WinXSPRO, all the files were generated in
Microsoft Excel with Position-Elevation
Free Form as the data format. In this paper,
the outputs related to stage (H), area (A),
wetted perimeter (P), and hydraulic radius
(R), were used in the further computations.
Development of a semi-
automated extension tool
A semi-automated extension tool
was developed in the Microsoft Excel
environment to enable the consistent
application of the various hydraulic exten-
sion methods with limited manual user
input. The semi-automated tools consist of
different functions (as applicable to each
hydraulic extension method) to ultimately
provide peak discharges at specific stages
above the structural limit.
SE and LE methods
The analyses were conducted and auto-
mated in two distinctive phases: (i) up to
structural limit, and (ii) above structural
limit. Rating curves, for the whole H-range
were generated automatically, while any
trend line evaluations were done manually
to determine the best-fit polynomial func-
tion based on the coefficient of determina-
tion (r 2) and the estimated discharge (QE).
In each case, the best trend line equation
was used for the required extension.
VE methods
All the VE methods shared an input
worksheet, with inputs for H, A, R, and v up
to structural limit and separate inputs for
above-structure-limit geometric proper-
ties. Essentially, first-order polynomial
equations were generated to enable the RC
fitting and subsequent extension of the
SD relationships for above-structure-limit
conditions.
SAM
Essentially, the SAM is not regarded as a
simplified computation where Manning’s
(MA) or Chézy’s (CHZ) equation for open-
channel flow is just used to determine
discharges at different stages. Hence, the
following steps are typically required
and were applied accordingly to deter-
mine the: (i) upstream and downstream
conveyance from known geometric and
hydrometric properties (using separate
worksheets for SAM-MA and SAM-CHZ),
(ii) average conveyance, (iii) energy slope
by assuming the velocity head equals
zero, and (iv) corresponding discharge
and assumed discharge. These steps were
automated in Microsoft Excel using Visual
Basic for Applications (VBA) and the
GOAL SEEK function.
SBA
The developed SBA worksheet is based on
the methodology proposed by Chow (1959),
i.e. the standard step method for natural
(non-prismatic) channels. The Froude

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 21
 of the estimate (SEE, Equation 1), mean
absolute relative error (MARE, Equation
Input SD relationship and/or Input SD relationship and/or
geometry up to structural limit geometry above structural limit 2), root mean square error (RMSE,
Equation 3), coefficient of determination
(r², Equation 4), Nash-Sutcliffe coefficient
(NSE, Equation 5), and the z-test (Equation
Output: SE: Review best-fit LE: Logarithmic
6), were used as assessment criteria to
Standard RC trend line transformation
compare and assess the indirect extension
methods’ results to those of the at-site
direct extension (benchmark) methods.
Output: Extend RC using the Extend RC using a The z-test indicates how many standard
SE RC best-fit trend line polynomial function
deviations a sample scores above or below
the mean, and the results are classified as:
(i) similar (0 ≤ z ≤ 2), (ii) marginally differ-
Output: VE-SA: Output: ent (2 < z ≤ 2.5), (iii) significantly different
VE-SA RC Compute/plot v vs H LE RC (2.5 < z ≤ 3), and (iv) unacceptable (z > 3)
(Williams 2023).
Each indirect hydraulic extension
Output: VE-HRA: SAM-MA/CHZ: method was ranked against the different
VE-HRA RC Compute/plot v vs R Run procedure GOF criteria and summed and/or averaged
to provide the overall performance. Finally,
the overall rankings and associated relative
frequency of the various z-test ranking
Output: VE-MA:
VE-MA RC Plot Q vs AR2/3 clusters – as defined in (i) to (iv) above –
were used to establish the hierarchical
Output:
order of the various indirect hydraulic
SAM-MA/CHZ RC
extension methods.

⎫
Output: SBA-MA/CHZ:

⎪N
⎪
SBA-MA/CHZ RC Trial-and-error process

⎪ ∑ (QEi – QE)2 –
(N – 2) ⎪⎪i=1
1
SEE = 

⎪
⎭
Figure 4 R
 C extension flow diagram applicable to indirect hydraulic extension methods
(Williams 2023)

⎫N ⎫ ⎫
⎪ ∑ (QBi – QB)(QEi – QE)⎪ ⎪
2 0.5

⎭i=1 ⎭ ⎪
⎪
number (FR) for each cross-section was Typically, the known water surface eleva-

⎪
⎪
determined to define the flow regime, i.e. tions (up- and downstream) were used as  (1)

⎪
N

⎭
super- and/or sub-critical flow. Essentially, the default boundary conditions at most of
∑ (QEi – QB)2
iterative trial-and-error calculations were the flow-gauging sites under consideration. i=1
used until channel profile convergence was In the absence of these known water surface
1 N |QEi – QBi|
reached, i.e. the profile approximates the elevations, the critical depth was entered
uniform depth profile and energy levels are as the downstream boundary condition,
MARE = 100
N
∑ QBi
(2)
i=1
in equilibrium. The procedure matrix or given that sub-critical flow conditions
flow diagram associated with all the indi- (FR < 1) were present at all the flow-gauging
N
rect hydraulic extension methods is shown sites under consideration. The simulated
∑ (QEi – QBi)2
in Figure 4. water surface levels were compared to the i=1
RMSE = (3)
observed water levels, while adjusting the N
1-D HEC-RAS steady flow modelling Manning’s n-values until similar water
The modelling in HEC-RAS was limited to surface levels were reached. N 2
steady flow analyses at the flood sections ∑ (QBi – QB)(QEi – QE)
i=1
only, given that the calibration data at the Assessment of results r² = (4)
N N
flow-gauging weirs is regarded as unsuitable A ranking-based selection procedure was
∑ (QBi – QB)2 ∑ (QEi – QE )2
for RC extensions at above-structure-limit developed to evaluate and select the best i=1 i=1
flow conditions. In addition, the calibration performing indirect hydraulic extension
data also lacks the information required methods at each flow-gauging site. In other N
to conduct an unsteady flow analysis words, the indirect extension method’s ∑ (QBi – QEi)2
i=1
for the full range of flows at the various estimation results (QE) were compared to NSE = 1 – (5)
N
flow-­gauging sites under consideration. the at-site or benchmark extension results
∑ (QBi – QB)2
The desired flow regime, i.e. sub-critical, (QB) at each site (i) by considering a large i=1
supercritical, and/or mixed flow, was set of quantitative GOF criteria (Zhong &
defined prior to the actual simulation run Dutta 2015) to assess each method’s accu- QBi – QEi
z=  (6)
by specifying the boundary conditions. racy and bias. Typically, the standard error √σBi + σEi

22 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
Where:
SEE is the standard error of estimate
(m3/s)
MARE is the mean absolute relative error
(%)
RMSE is the root mean square error
r² is the coefficient of determination
NSE is the Nash-Sutcliffe coefficient
z is the z-test score
N is the sample size
QBi is the at-site benchmark discharge
(m3/s)
QB is the mean of the at-site bench-
mark discharge (m3/s)
QEi is the estimated (indirect) discharge
using extension methods (m3/s)
QE is the mean of the estimated (indi-
rect) discharge (m3/s)
σBi is the standard deviation of the at- Figure 5 Flow-gauging weir X3H008 in the Sand River (Rademeyer 2023)
site benchmark discharge (m3/s)
σEi is the standard deviation of the esti- was used as the preferred SD extension Given that the SBA-CHZ method is used
mated (indirect) discharge (m3/s). method. Hence, the SBA was also used as the benchmark method (default ranking
as a benchmark against which the other = 1) at this site, the SBA-MA demonstrated
indirect extension methods were compared the second-best overall ranking, followed
RESULTS AND DISCUSSION up to a stage of 5.4 m and corresponding by the HEC-RAS (FS) modelling in the
In this section, the overall ranking results discharge of 2 972 m3/s. The most recent 3rd position. By considering the individual
associated with each flow-gauging site are cross-sectional survey at the flow-gauging GOF statistics, the latter SAM-CHZ and
presented in a summative format. However, weir is inclusive of five flood sections (FSs). SAM-MA methods also proved to be
given that all the indirect hydraulic exten- The current (existing) SD relationship table equally accurate by being interchangeably
sion methods and HEC-RAS modelling (DT10) applicable to X3H008 is listed in ranked at either the 4th or 5th position.
were applied at flow-gauging site X3H008, Table 2. Overall, the LE is the most inappropriate
the latter gauging site is discussed in detail The SD RC extension results using all method with the poorest ranking for all the
as a representative example. Thereafter, the the indirect hydraulic extension methods GOF statistics under consideration.
overall results and performance achieved and HEC-RAS modelling are shown in Overall, the hydraulically correct
at the 10 flow-gauging sites are discussed Figure 6 (on page 24). The GOF statistics methods resulted in the lowest SE E ,
collectively. applicable to each method are listed in MARE, RMSE, and z-test values. All the
Table 3 (on page 24). methods demonstrated a high degree of
Flow-gauging weir X3H008 In Table 3, the various indirect SD association, with r2 values > 0.99; however,
Flow-gauging weir X3H008 is located in extension methods, except for the SE, this only highlights a high correlative
the Sand River, Mpumalanga Province. LE, and SAM-MA methods, provided trend between the estimated QEi values
The weir was constructed in 1967. As estimated discharge (QEi) values that as suggested by the various extension
shown in Figure 5, the structure consists exceeded the at-site benchmark discharges methods and the benchmark QBi values.
of a sharp-crest, hydro-flume, and broad- (QBi). The latter underestimations using Hence, the NSE results should also be
crested flank walls (Mathole 1999). The the SE and LE methods were evident for considered, whereas the VE-HRA and
structural limit is at a stage of 1.95 m, with stages (H) > 2 m (65% ≤ MARE ≤ 70%), VE-MA methods demonstrated the
an associated discharge of 76 m3/s. In the while the SAM-MA underestimations were lowest NSE values of ± 0.19 despite
latest DWS Calibration Report, the SBA much less significant, with MARE = 1.5%. their r2 values ≈ 0.99.

Table 2 Estimated discharge table 10 (DT10) at X3H008 based on the SBA (Nathanael et al 2018)
Stage increments (m) and discharge (m3/s)
Stage (m)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

0 0.0 0.2 0.6 1.1 1.8 2.8 4.7 7.1 9.9 13.0

1 16.5 20.3 24.4 28.7 34.1 39.1 45.4 52.7 61.0 70.2

2 84.7 107.6 137.0 172.4 213.8 261.2 315.2 372.1 429.1 488.9

3 552.8 618.7 688.0 760.3 836 914 995 1 080 1 167 1 258

4 1 351 1 448 1 548 1 650 1 756 1 865 1 976 2 091 2 209 2 330

5 2 453 2 580 2 710 2 843 2 972 – – – – –

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 23
 3 000
Benchmark
DT 8
DT 9
DT 10
2 500 SE
LE
VE-SA
VE-HRA
VE-MA
2 000 SAM-MA
SAM-CHZ
SBA-MA
Discharge (m3/s)

SBA-CHZ
HEC-RAS (FS)
1 500

1 000

500

0
0 1 2 3 4 5
Stage (m)

Figure 6 Indirect SD RC extensions in comparison to the benchmark rating at X3H008 (Williams 2023)

Table 3 GOF results at X3H008 (Williams 2023)

VE SAM *SBA HEC-RAS
Criteria SE LE
SA HRA MA MA CHZ MA CHZ (FS)

Sample size (N) 35 35 35 35 35 35 35 35 35 35

Average 319 271 1348 1809 1839 1217 1273 1235 1235 1236

Standard deviation 174 146 899 1436 1414 877 917 890 890 891

SEE (Eq (1); m3/s) 5.7 4.4 31.2 66.3 27.0 0.0 0.2 0.0 0.0 0.0

SEE ranking 7 6 9 10 8 3 5 2 1 4

MARE (Eq (2); %) 65.4 70.2 16.3 35.0 39.7 1.5 3.0 0.0 0.0 0.1

MARE ranking 9 10 6 7 8 4 5 2 1 3

RMSE (Eq (3)) 1156.5 1211.5 116.8 788.4 794.7 22.8 0.0 0.0 0.0 1.3

RMSE ranking 9 10 6 7 8 5 3 2 1 4

r 2 (Eq (4)) 0.999 0.999 0.999 0.998 1.000 1.000 1.000 1.000 1.000 1.000

r 2 ranking 8 7 9 10 6 4 5 2 1 3

NSE (Eq (5)) –0.74 –0.91 0.98 0.19 0.18 1.00 1.00 1.00 1.00 1.00

NSE ranking 9 10 6 7 8 5 3 2 1 4

z-test (Eq (6)) 28.1 30.0 2.7 11.9 12.6 0.4 0.9 0.0 0.0 0.0

z-test ranking 9 10 6 7 8 4 5 2 1 3

Sum of rankings 51 53 42 48 46 25 26 12 6 21

Overall ranking 9 10 6 8 7 4 5 2 1 3

*Benchmark method, MA (Manning’s approach), CHZ (Chézy’s approach), and FS (flood section)

24 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
Table 4 Summary of the GOF-based rankings at the 10 flow-gauging sites (Williams 2023)
VE SAM SBA HEC-RAS
Gauging site SE LE
SA HRA MA MA CHZ MA CHZ (FS)

A4H005 4 5 3 1 2 - - - - -

A6H035 4 1 2 3 5 - - - - -

C2H003 3 1 5 4 2 - - - - -

C5H014 10 9 7 8 6 4 5 2 1 3

H5H004 10 9 7 8 6 4 5 2 1 3

J1H018 9 8 6 7 10 4 3 2 1 5

J4H002 7 9 5 8 10 6 4 2 1 3

K2H002 1 2 3 4 5 - - - - -

V1H026 1 2 3 5 4 - - - - -

X3H008 9 10 6 8 7 4 5 2 1 3

Avg rankings 5.8 5.6 4.7 5.6 5.7 4.4 4.4 2 1.2 3.4

Overall ranking 10 8 6 7 9 5 4 2 1 3

Summary of results
The overall ranking results, as listed in
Table 4, highlight that the SBA, SAM,
and 1-D HEC-RAS steady flow modelling
performed the best. The other indirect
extension methods were characterised by
larger statistical differences between the
at-site benchmark values (QBi) and the
modelled values (QEi). The VE-MA and SE
methods are regarded as the least appropri-
ate methods, ranked respectively in the 9th
and 10th positions.
SE performance
The SE method is ranked overall in the
10th position. The method tends to either
over- or underestimate the at-site bench-
mark discharges (QBi). At K2H002 and
V1H026 it provided satisfactory results due
to no changes in the geometry or hydraulic
conditions above the structural limit. This
method is not suited to extend SD relation-
ships at sites with complex hydraulics or
geometry, and the use thereof should be
limited to flow-gauging sites with adequate
data and reasonably simple site charac-
teristics and hydraulics. It demonstrated
the poorest performances at C5H014 and
H5H004.
LE performance
The LE method is ranked overall in the 8th
position. The method tends to either over-
or underestimate the at-site benchmark
discharges (QBi). At A6H035 there are no
apparent transitions beyond or above the
structural limit; hence, the subsequent
number one ranking at this site. The LE
method performed marginally better than
the SE method at all the flow-gauging sites
under consideration, except at A4H005,
J4H002, K2H002, V1H026, and X3H008.
However, this method is not recom-
mended for significant extensions above
the structural limit; typically, in these
above-structure-limit ranges, it proved
to result in estimated discharges (QEi)
significantly different from the benchmark
discharges (QBi).
VE performance
The VE-SA (6th position) and VE-HRA
(7th position) methods performed better
than the VE-MA (9th position) method;
however, the application of all these meth-
ods was limited by data availability. The VE
methods require a complete cross-section
of the weir at high flows, and such data
sets were not available at most of the sites
under consideration, except at A4H005
where these methods were ranked as the
top three methods. Hence, based on the
promising results achieved at some of
the flow-gauging sites, the VE methods
should be considered more often when SD
relationships need to be extended. The
VE methods are also simpler to apply as
opposed to the often-preferred SAM and
SBA methods.
SAM performance
The SAM-CHZ is ranked overall in the
4th position, while the SAM-MA is ranked
in the 5th position. In comparison to the
SE, LE, and VE methods, it is hydraulically
more correct. At J1H018 and J4H002, the
SAM-CHZ proved to be hydraulically more
correct (3rd to 4th position) and agreed
well with the SBA. By considering the
individual GOF statistics, the SAM-CHZ
and SAM-MA methods also proved to
be equally accurate at flow-gauging sites
C5H014, H5H004, and X3H008 by being
interchangeably ranked at either the 4th or
5th position.
SBA performance
The SBA-CHZ is ranked overall in the 1st
position, while the SBA-MA is ranked in
the 2nd position. However, the SBA was
also used as the preferred benchmark
method at 50% of the flow-gauging sites
under consideration. The latter preferential
use and associated rankings were also
confirmed by Roux et al (2015): “From
experience, DWS Flood Studies regard the
SBA method as superior to the SAM when
flood peaks need to be estimated.” The
latter statement is also supported in inter-
national literature. The SBA is also less
affected by hydraulic conditions than the
SAM, and is therefore regarded as a more
versatile method.
The Chézy’s absolute roughness (ks)
version of the SBA, i.e. SBA-CHZ, is
preferred to the Manning’s n-value version
(SBA-MA). This is ascribed to the fact that
Manning’s n-value is expressed in non-
measurable units, i.e. s/m1/3, and is con-
stant for a given discharge associated with
a specific velocity and hydraulic radius.
The Chézy equation makes allowance
for changes in the roughness coefficient
C (m0.5/s), as changes in the flow depth
and subsequently the hydraulic radius (R)
occur. As a result, channel profile conver-
gence is normally achieved sooner when

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 25
Chézy’s C-value (function of ks and R) is
used in the SBA. Manning’s n-value does
vary with changes in the hydraulic radius;
however, there is no explicit mathematical
procedure to determine this. Additionally,
it is also easier to visualise Chézy’s absolute
roughness (ks) as it represents the sizes
of irregularities on the channel beds and
sides, expressed in metres.
It is also important to note that the
starting water surface level or elevation
in the SBA is regarded as unknown or
indefinite; hence, several backwater profiles
based on different arbitrary water surface
levels should be computed until conver-
gence occurs, i.e. the water levels cor-
respond with the uniform flow profile and
energy levels are in equilibrium. Therefore,
if convergence is not reached (probably
owing to insufficient cross-sectional survey
data), additional and more distant (up- and/
or downstream) cross-sections would be
required. At some of the flow-gauging sites
under consideration, DWS had already
established an arbitrary datum (reference
level), and known water surface levels (e.g.
flood marks). Subsequently, defining an
arbitrary point to reach profile convergence
was not required in these cases.
In assigning roughness parameters,
i.e. Manning’s n-values, it should not be
limited to a cross-section, but it should
rather be representative of the entire river
reach under consideration (George et al
1989). Observations made during this
study confirmed the difficulty in assigning
Manning’s n-values for a river reach and
the cross-sections under consideration.
Hence, the combined influence of various
input parameters (e.g. flow depth, turbu-
lence, bed forms, vegetation, sedimenta-
tion and bed load, debris, obstructions,
cross-section shape, geomorphological
processes, and floodplain-main channel
interactions) needs to be considered when
assigning n-values during high flows
(Lumbroso & Gaume 2012). Irrespective of
the method used – e.g. SAM, SBA and/or
HEC-RAS – if representative and correct
n-values are assigned, comparable results
will be achieved.
HEC-RAS performance
It was evident that the HEC-RAS results
improved as the number of more detailed
and distant cross-sections increased.
Therefore, by having a continuous surface,
i.e. digital terrain model, from which high-
accuracy cross-sections can be extracted,
the modelling results can be improved
26
significantly. Typically, owing to insuf-
ficient data being available, no HEC-RAS
modelling was possible at flow-gauging
sites A4H005, A6H035, C2H003, K2H002
and V1H026. At the latter flow-gauging
sites, only calibration data up to the
structural limit was available, subsequently
resulting in discontinued flow at above-
structure-limit conditions. In applying
HEC-RAS at data deficient flow-gauging
sites, overestimations will typically occur
up to the structural limit, while above-
structure-limit flows will either be under-
or overestimated.
CONCLUSIONS
The overall aim of this paper was to assess
a selection of indirect hydraulic exten-
sion methods against direct extension
(benchmark) methods at selected flow-
gauging sites in South Africa. In terms of
the benchmark methods, the following
approaches were used at the various flow-
gauging sites: (i) dam spillage (K2H002),
(ii) hydrograph analysis (V1H026), (iii) level
pool (back) routing (A4H005, A6H035, and
C2H130), and (iv) SBA (C5H014, H5H004,
J1H018, J4H002, and X3H008).
The SBA-CHZ was ranked overall in
the 1st position, while the SBA-MA was
ranked in the 2nd position. The SBA was
used as the preferred benchmark method
at 50% of the flow-gauging sites under
consideration, given that the SBA is less
affected by hydraulic conditions than the
SAM and is therefore regarded as a more
versatile method. The HEC-RAS steady
flow modelling was ranked overall in the
3rd position. In principle, the HEC-RAS
steady flow modelling is also based on the
SBA; hence, the latter performance was
expected. However, the steady flow model-
ling was in general limited by data suitabil-
ity and availability. It was evident that the
HEC-RAS results improved as the number
of more detailed and distant cross-sections
increased. The SAM-CHZ and SAM-MA
were ranked overall in the 4th and 5th
position, respectively. In comparison to the
SE, LE, and VE methods, the SAM proved
to be hydraulically more correct. However,
the performance of the SAM is highly
dependent on detailed cross-sections, while
roughness coefficients for between-reaches
are also required.
The SBA, SAM, and 1-D HEC-RAS
steady flow modelling are therefore
regarded as the most appropriate indirect
estimation methods to reflect the hydraulic
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
conditions during high discharges at a
flow-gauging site. In general, any extension
method must be hydraulically correct if
it is to be used as a robust approach to
extend SD RCs beyond the structural limit.
The extension of an RC is significantly
more affected by the site and river reach
geometry, initial hydraulic conditions,
flow regimes, and level of submergence at
high discharges, than the actual extension
method used.
Hence, there is no one-size-fits-all
approach available for the extension of SD
RCs in South Africa. For the interim, as rec-
ommended by Gericke and Smithers (2017),
individual stage extensions (HE ≤ 1.2 H
subjected to QDE ≤ 0.05 QDxi) should be
considered for bankfull or above bankfull
flow conditions in cases where site-specific
data is unavailable. It is also recommended
that the river geometry for a large sample of
representative flow-gaugings in South Africa
be investigated to establish which simplified
profile, e.g. trapezoidal cross-section with
1:2 side slopes, could render acceptable dis-
charges for bankfull and/or above bankfull
flow conditions.
ACKNOWLEDGEMENTS
Support for this research by the Central
University of Technology Free State (CUT)
and the Water Research Commission
(WRC) is gratefully acknowledged. We
also wish to thank Mr Danie van der Spuy
(former Specialist Engineer, Directorate
National Hydrological Services: Flood
Studies) for his expert and technical
advice pertaining to the extension of
stage-discharge relationships, as well as the
anonymous reviewers for their constructive
review comments which have helped to
significantly improve the paper.
REFERENCES
BSI ISO (British Standards Institution & International
Organisation for Standardisation) 1998. BSI ISO
1100-2. Measurement of Liquid Flow in Open
Channels: Determination of the Stage-Discharge
Relation. London: BSI.
Chow, V T 1959. Open-channel Hydraulics. New York:
McGraw-Hill.
Dymond, J R & Christian, R 1982. Accuracy of
discharge determined from a rating curve.
Hydrological Sciences Journal, 27(4): 493–504.
doi: 10.1080/02626668209491128.
EA UK (Environmental Agency United Kingdom)
2002. Stage 1: Extension of rating curves at gauging
stations. Report No 1. Bristol: EA UK.
EA UK 2003. Stage 2: Extension of rating curves at
gauging stations using simple hydraulic techniques.
Report No 2. Bristol: EA UK.
George, J, Arcement, J R & Schneider, V R 1989. Guide
for selecting Manning’s roughness coefficients for
natural and flood plains. Water Supply Paper 2339.
US Geological Survey. Denver. CO: Department of
the Interior. Available at: https://pubs.usgs.gov/
wsp/2339/report.pdf.
Gericke, O J & Smithers, J C 2017. Direct estimation
of catchment response time parameters in medium
to large catchments using observed streamflow
data. Hydrological Processes, 31(5):1125–1143.
doi: 10.1002/hyp.11102.
Hardy, T, Panja, P & Mathias, P 2005. WinXSPRO: A
channel cross-section analyser. Lakewood, CO: US
Department of Agriculture, Forest Service.
Lang, M, Pobanz, K, Renard, B, Renouf, E & Sauquet,
E 2010. Extension of rating curves by hydraulic
modelling with application to flood frequency analysis.
Hydrological Sciences Journal, 55(6): 883–898. https://
doi.org/10.1080/02626667.2010.504186.
Lindner, G A & Miller, A J 2012. Numerical
modelling of stage-discharge relationships in
urban streams Journal of Hydrological Engineering,
17: 590–596.
Lumbroso, D & Gaume, E 2012. Reducing the
uncertainty in indirect estimates of extreme
flash flood discharges. Journal of Hydrology,
414–415: 16–30. https://doi.org/10.1016/j.
jhydrol.2011.08.048.
Journal of the South African Institution of Civil Engineering
Maeder, C 2015. The road to HEC-RAS. Middleton,
WI: Civil GEO, US. Available at: https://www.
civilgeo.com/blog/the-road-to-HEC-RAS.
Mathole, J 1999. Kalibrasieversoek (X3H008). Report
No DT5. Pretoria: Department of Water Affairs and
Forestry.
Menné, T C 1960. The role of hydrology in planning the
development of a country. Hydrological Research
Technical Report No 9. Pretoria: Department of
Water Affairs.
Nathanael, J J 2016. Assessing the performance of
regional flood frequency analysis methods in
South Africa. MSc Dissertation. Pietermaritzburg:
University of KwaZulu-Natal.
Nathanael, J J, Rademeyer, P F & Van der Spuy, D 2018.
Discharge rating improvement of rating table and
high flow data Sand River at Exeter. Report No
X3H008DT-2018.12. Pretoria: Department of Water
and Sanitation.
Petersen-Øverleir, A & Reitan, T 2009. Accounting
for rating curve imprecision in flood frequency
analysis using likelihood-based methods. Journal of
Hydrology, 366(1–4): 89–100.
Rademeyer, P 2023. Personal communication. Pretoria:
Directorate: National Hydrological Services,
Department of Water and Sanitation.
Ramsbottom, D M & Whitlow, C D 2003. Extension of
rating curves at gauging stations: Best practice guidance
manual. Bristol, UK: Environmental Agency.
Roux M, Rademeyer, P F & Van der Spuy, D 2015.
Discharge rating improvement of rating table for
Volume 65 Number 2 June 2023
Riet River at gauging weir C5H014. Report No
C500-H014-2015.03. Pretoria: Department of Water
and Sanitation.
Shao, Q, Dutta, D, Karim, F & Petheram, C 2018. A
method for extending stage-discharge relationships
using a hydrodynamic model and quantifying
the associated uncertainty. Journal of Hydrology,
556(2018): 154–172.
Smithers, J C 2022. The initiation of a National Flood
Studies Programme for South Africa: Overview
and progress to date. Paper presented at the IAHS
2022 11th Scientific Assembly. Montpellier, France.
doi: 10.5194/iahs2022-229.
USACE (United States Army Corps of Engineers)
2016. HEC-RAS river analysis system user’s manual.
Davis, CA: Hydrologic Engineering Centre.
Wessels, P & Rooseboom, A 2009. Flow-gauging
structures in South African rivers. Part 1: An
overview. Water SA, 35(1): 1–10.
Williams, V H 2023. Assessment of indirect estimation
methods to extend observed stage-discharge
relationships for above-structure-limit conditions
at flow-gauging weirs. MEng Dissertation.
Bloemfontein: Central University of Technology,
Free State.
Zhong, X & Dutta, U 2015. Engaging Nash-Sutcliffe
efficiency and Model Efficiency Factor Indicators
in selecting and validating effective light rail
system operation and maintenance cost models.
Journal of Traffic & Transportation Engineering,
3(2015): 255–265.
27
TECHNICAL PAPER
Advanced numerical
JOURNAL OF THE SOUTH AFRICAN
INSTITUTION OF CIVIL ENGINEERING modelling of the nonlinear
mechanical behaviour
ISSN 1021-2019 (print) | ISSN 2309-8775 (online)
Vol 65 No 2, June 2023, Pages 28–38, Paper 1492

KEVIN BRAUN received his BEng and BEngHons

at the University of Pretoria and is currently of a laterally loaded
pile embedded in stiff
completing his MEng in Civil Engineering
(Structures). His research work consists of
modelling multiple soil-structure interaction

unsaturated clay
models of a singular pile embedded in soil,
where formulas using AI (artificial intelligence)
and machine learning algorithms will be able to
predict the failure force and horizontal displacement of a laterally loaded pile.
Contact details:
Department of Civil Engineering, University of Pretoria
Private Bag X20, Hatfield, Pretoria 0028, South Africa K T Braun, N Bakas, G Markou, S W Jacobsz
E: u17031215@tuks.co.za

DR NIKOLAOS BAKAS is a Senior Data Scientist

at the National Infrastructures for Research
and Technology, GRNET. He has been a
lecturer at Neapolis University Pafos, and an
Associate Research Scientist at the
Computation-based Science and Technology
Research Center, at The Cyprus Institute. His
research interests comprise the fundamental
mathematical modelling of machine learning algorithms, as well as
applications in a variety of thematic areas. Apart from research, he has been
a PI for multiple industrial applications of AI (artificial intelligence).
Contact details:
National Infrastructures for Research and Technology – GRNET
7 Kifisias Avenue, Athens 11523, Greece
E: nibas@grnet.gr
PROF GEORGE MARKOU received his PhD
degree at the National Technical University of
Athens, Greece. His speciality involves
numerous scientific fields related to the
modelling of structures through the use of the
finite element method. His recent research
work addresses the modelling of composite
structures under seismic loading conditions,
soil-structure interaction problems, assessment of retrofitted structures,
smart monitoring of structures, and the use of machine learning algorithms
for solving engineering problems. Prof Markou is also using
high-performance computing in the development of predictive models
through machine learning algorithms. He is currently an Associate Professor
at the University of Pretoria in the Department of Civil Engineering.
Contact details:
Department of Civil Engineering, University of Pretoria
Private Bag X20, Hatfield, Pretoria 0028, South Africa
E: george.markou@up.ac.za
PROF SW JACOBSZ completed BEng and
MEng degrees at the University of Pretoria,
and a PhD at the University of Cambridge. He
was employed by Jones & Wagener
Consulting Engineers as geotechnical
engineer for ten years, and was then
appointed as Associate Professor in the
Department of Civil Engineering at the
University of Pretoria in 2010. His research interests include physical
modelling of geotechnical and soil-structure interaction problems and
aspects related to unsaturated soil mechanics. He has been involved in
several large-scale field trials, including large surcharge trials on residual
dolomite, large-scale testing of piles in expansive clays and monitoring the
in-situ state on a range of tailings dams
Contact details:
Department of Civil Engineering, University of Pretoria
Private Bag X20, Hatfield, Pretoria 0028, South Africa
E: sw.jacobsz@up.ac.za
Capturing and understanding the ultimate limit state behaviour of reinforced concrete piles
embedded in soil requires the use of advanced tools or the performance of expensive tests. An
experiment was performed where reinforced concrete piles embedded in a stiff unsaturated
clay profile were load-tested on-site. However, even though in-situ experiments can provide
engineers with valuable insight, their cost and time limitations come with restrictions, especially
when dealing with a parametric investigation on the soil’s material properties, the size of the
piles, or the piles’ material properties. The objective of this research work was to numerically
model the nonlinear mechanical behaviour of laterally loaded full-scale piles through detailed
3D modelling, and perform an in-depth parametric investigation to provide answers to
unknown factors that the actual physical experiment could not answer. Furthermore, this
work serves as a pilot project that will be used to pave the way in developing multiple soil-
structure interaction models that will be used to generate a dataset that helps the creation of
predictive models through machine learning algorithms. For the needs of this research work,
the reinforced concrete piles were discretised with 8-noded isoparametric hexahedral elements
that accounted for cracking through the smeared crack approach. Steel reinforcement bars and
stirrups were simulated as embedded rebar elements, while the soil domain was also discretised
through 8-noded hexahedral elements. Most of the required material properties assumed
during the nonlinear analyses were defined according to relevant laboratory experiments.
According to the numerical investigation, it was found that the proposed numerical model has
the ability to reproduce the experimental results with high accuracy, while providing in-depth
insight on the failure mechanisms for both the soil and reinforced concrete domains.
INTRODUCTION a lateral load analysis is used to determine
Piled foundations are commonly used the flexural behaviour. According to
to support structures in loose or soft Červenka and Papanikolaou (2008) the
soils, as shallow foundations are known methods to study the behaviour of piles
for their inability to avoid extreme set- subjected to lateral loading can be catego-
tlements and shear failure when used in rised into:
such soil domains. Piles can either be 1. limit state methods
subjected to vertical loads, lateral loads, 2. subgrade reaction methods
or a combination of vertical and lateral 3. elastic continuum approaches
loads. Karthigeyan et al (2006) mentioned 4. p-y methods
that it is complex to analyse piles under 5. finite element methods.
combined loading, and therefore the load In soil modelling, the combination of
cases should be modelled independently. A elasticity, plasticity and friction are options
vertical load analysis is used to determine when modelling soil behaviour. Atkinson
the bearing capacity and settlement, while (2017) states that these combinations can

Keywords: soil-structure interaction, reinforced concrete pile, soil Braun KT, Bakas N, Markou G, Jacobsz SW. Advanced numerical modelling of the nonlinear mechanical behaviour of a laterally loaded pile
embedded in stiff unsaturated clay. J. S. Afr. Inst. Civ. Eng. 2023:65(2), Art. #1492, 11 pages. http://dx.doi.org/10.17159/2309-8775/2023/v65n2a4
domain, mechanical behaviour, finite element method
The Journal of the South African Institution of Civil Engineering, which is distributed internationally, is a peer-reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND). Readers may therefore freely use and share the content as long as they credit the original creator and publisher, do
28 not change the material in any way, and do not use it commercially. Copyright of this article remains with the authors.
 σ’a
F Y
σ’f
G
σ’g
Y F
σ’y
εp εe
R B
0 εa
εp
(a)
Figure 1 S train hardening (a) and strain softening (b) (adapted from Atkinson 2017)
be combined into a single model known as
the critical state model. Atkinson (2017)
proves that by considering the stress-strain
relationship of soils, they can typically
behave in three different ways, namely:
elastic-perfectly plastic, elastic-plastic
hardening, and elastic-plastic softening. A
perfectly plastic material continues to strain
with constant stress, and therefore plastic
deformation occurs during plastic deforma-
tion only. Hence, it does not experience
elastic deformation under plastic strain. An
elastic-plastic material that strain hardens
experiences elastic strain during plastic
deformation, while an elastic-plastic mate-
rial that strain softens releases elastic strain
during plastic deformation (Atkinson 2017).
Figure 1 represents the stress-strain curves
for an elastic-plastic strain hardening and
softening material, respectively.
Jardine et al (1986) studied the influ-
ence of non-linear stress-strain experi-
enced within soil-structure interaction
(SSI) problems. It was stated that from
field and laboratory studies, at very small
strains, most soils experience a non-linear
stress-strain behaviour. It was therefore
concluded that, despite the treatment of
soil as a linear elastic material in SSI prob-
lems, which represents an effective tool for
expressing soil stiffness if non-linearities
are not considered, the results can be mis-
leading. Breysse et al (2005) investigated
the effect of soil heterogeneity within SSI
problems. Breysse et al (2005) showed that
soil heterogeneity induces changes in the
differential settlement, bending moments,
stresses, and possibly cracking within the
SSI problem which cannot be predicted if
Journal of the South African Institution of Civil Engineering
σ’a
σ’y
σ’g
σ’f
0 εa
εp
homogeneous conditions are assumed. Soil
heterogeneity also induces stresses which
are dependent on the soil variability, the
soil-structure stiffness ratio, and the soil-
structure length ratio (soil fluctuation scale
and structural dimensions).
According to Kausel (2010), research
work considering the early history of SSI
describes the SSI phenomenon as some-
what difficult to define. This was because
of the complex dynamic interactions due to
the amplification of seismic waves present
within the soil layers. Another research
work analysed the non-linear behaviour of
laterally loaded piles (Kavitha et al 2016),
where they investigated the soil-pile inter-
action effect when loaded laterally. It was
reported that the soil-pile interaction effect
depended on the soil properties, loading
type, ground slope, as well as the pile mate-
rial and diameter (Kavitha et al 2016).
A study by Gravett and Markou (2021)
performed a state-of-the-art investigation
on wind turbine structures founded on soft
clay, where SSI was also considered. This
work investigated the interaction of soil
and structural mechanics, and foresaw the
derivation of an optimum pile foundation
geometry based on modal and ultimate
limit nonlinear pushover analyses. Gravett
and Markou (2021) used ReconanFEA
(2020) to perform the analyses in a detailed
3D approach to discretise, model and ana-
lyse numerous finite element (FE) models.
They reported that the use of hexahedral
finite elements was ideal for this type of
SSI problem. It is noteworthy to refer to
the recent research work by Gravett et al
(2021), where the fundamental period of
Volume 65 Number 2 June 2023
G
–εe
εp
(b)
reinforced concrete (RC) buildings was
investigated, which included SSI effects. In
their research work, a digitally generated
dataset was used to derive a closed-form
formula that was able to accurately predict
the fundamental period of RC buildings
with and without accounting for the SSI
effect. A similar approach that will foresee
the use of artificial intelligence (AI) and
machine learning algorithms will be used
by the authors after the successful delivery
of the research work discussed herein.
The finite element method (FEM) is
one of the most commonly used numeri-
cal approaches in industry, as well as for
research purposes. This method aims
to predict the mechanical behaviour of
structures, allowing and assisting engineers
to numerically investigate structures.
For this reason, the FEM-based software
(ReconanFEA 2020) is used for the needs
of this research work. Furthermore, by
analysing numerical models under differ-
ent loading conditions and using a variety
of meshes, it can be determined if the
structural design is safe or not (Markou &
Roeloffze 2021). Modelling of piles embed-
ded in soil was performed by Gravett and
Markou (2021), where experimental and
numerical results showed that ReconanFEA
(2020) has the ability to objectively capture
the mechanical response of this complex
problem. Therefore, the modelling that was
conducted herein to investigate the behav-
iour of piles subjected to different loading
conditions follows a similar approach,
adopting detailed 3D modelling of solid
elements to discretise both the concrete
and soil domains.
29
 A research project which aimed to inves-
tigate the behaviour of piled foundations, σ2
particularly for wind turbines in partially
saturated clay soils, was recently undertaken
in South Africa. Expansive soils are located σy
in many sub-arid and sub-humid parts
across the African continent, and therefore
the design of superstructures within or on
expansive soils is critical for the future of
the socio-economy of Africa. Field tests
were carried out on a site near Vredefort
in South Africa. This site was underlain by
σ1
expansive clay which extends to depths in
excess of 7 m, with bedrock occurring at –σy σy
about 15 m. Two sets of lateral pile load tests
were conducted with the soil in a dry (in-
situ) state, as well as after saturating the soil.
The objective of this study was to use
advanced numerical models to analyse the
nonlinear mechanical behaviour of RC piles
–σy
embedded in clay by replicating in-situ lateral
load tests. Concrete, steel and soil properties
were modelled in the numerical analysis per-
formed herein according to the material tests
performed for the purposes of the physical Figure 2 V
 on Mises 2D yielding criterion curve (adapted from Gravett & Markou 2021)
experiment. However, for unknown proper-
ties, specific engineering judgement had by Gravett and Markou (2021) describes The formulation that describes the
to be used to obtain representative values, how the corresponding hydrostatic and strength envelope of concrete (Markou et
especially in the case of the soil. Thereafter, deviatoric strains (ε 0 , γ 0) are used to pro- al 2018) is expressed as:
a sensitivity analysis was conducted on these duce the hydrostatic and deviatoric stress-
2 – τ 2 )cos θ + τ (2τ – τ )
2τ0c(τ0c
material properties to determine how sensi- es, by using a material model proposed by 0e 0c 0e 0c
2 – τ 2 )cos2θ + 5τ 2 – 4τ 2 τ 2
√4(τ0c
tive the assumptions of property values were Kotsovos and Pavlovic (1995) and extended 0e 0e 0c 0e
τ0u = (4)
in terms of the analysis. For this purpose, by Markou and Papadrakakis (2013). 2 – τ 2 )cos2θ + (2τ – τ )2
4(τ0c 0e 0e 0c
numerical software packages such as FEMAP
(Ouanta et al 2016) were used. By develop- σ 0 + σ id τ0 This expression represents a smooth con-
ε 0 = ε 0h + ε 0d = , γ0 = (1)
ing models through FEMAP and analysing 3KS 2GS vex curve, where θ is the rotational angle
them through the use of the ReconanFEA between the deviatoric stress vector and one
software package, while considering the input In Equation 1, σid expressed as a function of the projected stress principal axis located
parameters measured from the full-scale RC of (σ 0 , τ0 , fc) is an equivalent external stress, on the deviatoric plane. In addition, once the
pile load research project in Vredefort, South which accounts for the coupling effect failure criteria is satisfied, the smeared crack
Africa, correlations between the models between τ0 and ε 0d (Gravett & Markou approach is imitated where macro-cracking
and real-life experiments were performed. 2021). In addition, the bulk modulus KS (σ 0 , occurs (Gravett & Markou 2021).
Additionally, even though the physical τ0) and the shear modulus GS (σ 0 , τ0) are In terms of reinforcement modelling,
experiment did not proceed to ultimate obtained by ignoring σid. By considering the Menegotto-Pinto method, which
failure, it is shown herein that the numerical the relationship in Equation 1, the strains in also considers the Bauschinger effect,
simulations provided much missing informa- global coordinates are expressed as follows: is integrated within the ReconanFEA
tion, complementing the physical experiment software (Menegotto 1973). The stress-
by answering questions such as revealing the σ ij + σ idδij 3vS strain relationship for the steel material is
εij = – (σ 0 + σ id)δij(2)
failure mechanism, as well as the ultimate 2GS ES presented as:
lateral load at which the RC pile failed.
Where: δij is the Kronecker delta, σ ij and εij (1 – b)ε*
σ* = bε* + (5)
are the total stresses and strains, respec- (1 + ε*R)1/R
MODELLING OF CONCRETE tively. ES (σ 0 , τ 0 , fc) and νS (σ 0 , τ 0 , fc) are the
AND SOIL MATERIAL secant Young modulus and Poisson’s ratio, Where:
Modelling in 3D considers the effect of respectively, which are derived from KS and
out-of-plane stresses, where the stress- GS (Markou & Roeloffze 2021). ES and νS ε – εr
ε* = ,(6)
strain relationship for each stress-state is are expressed as: ε 0 – εr
expressed by the hydrostatic (σ 0) and devia-
toric components (τ 0) (Markou & Roeloffze 9KSGS 3KS – 2GS σ – σr
ES = , vS = (3) σ* = ,(7)
2021). The recent research work presented 3KS + GS 6KS + 2GS σ0 – σr

30 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering

Soil domain
Applied load
Figure 3 F inite element mesh of model 1 (left), load application (middle) and reinforcement
details (right)
3m
7m
8m
Soil
domain
10 m
Figure 4 F inite element mesh of model 2; the interface of the pile and soil is detached
R0 – a1ξ
R= ,
α 2 + ξ (8)
Where: b is the strain hardening ratio, and
ξ is the normalised plastic strain param-
eter. Furthermore, the parameters R0, a1
and a2 are assumed parameters depending
on the investigation of the problem at hand
(Gravett & Markou 2021).
For the case of the soil domain that was
discretised through the use of hexahedral
elements, the adoption of a nonlinear elas-
toplastic 3D material model was performed
that used a Von Mises yielding criterion
described in Equation 9. The numerical
investigation performed and presented in
this research work foresaw that the RC
pile is founded on unsaturated clay soil.
It is also important to note here that the
soil material model requires the definition
Table 1 Soil input parameters
Parameter
Young’s modulus
Compressive strength
Poisson’s ratio
Journal of the South African Institution of Civil Engineering
Concrete pile
Rebars and stirrups
Soil gap
600 mm
of only three material properties – the
E-value, Poisson’s ratio, and the uniaxial
ultimate strength. Figure 2 shows a typical
Von Mises yielding criterion curve.
(σ 1 – σ 3)2 – (σ 1 – σ 3)(σ 2 – σ 3) + (σ 2 –σ 3)2
= σ y2(9)
SOIL STRUCTURE
INTERACTION MODEL
The RC pile geometry used in the numeri-
cal analysis was developed based on the
design of the field piles, where piles of
600 mm in diameter with lengths of 7 m
were constructed. The pile was auto-
discretised using hexahedral elements
with an element size of 0.1 m. The soil sur-
rounding the pile was modelled to a radius
of 5 m and a depth of 8 m, while also being
Value Units
50, 100, 150 MPa
89, 100, 110, 120, 130, 140, 150 kPa
0.3 –
Volume 65 Number 2 June 2023
discretised with 8-noded isoparametric
hexahedral elements. This first model (see
Figure 3) assumed a full bond between
the concrete-soil interface. The reinforce-
ment of the pile (12Ø25 longitudinal and
Ø10@150 stirrups) was also discretised and
analysed in this study through the use of
embedded rod elements that can be seen in
Figure 3.
Given that the experiment foresaw the
application of static loads, the soil domain
was fixed at the bottom, and the nodes
along the perimeter of the soil domain
were restrained from displacing horizon-
tally, while allowed to undergo settlement
(z-global axis). Figure 3 also shows the
point of load application that was chosen
according to the field experiment (0.5 m
above the soil surface). The total applied
load within the model was divided into
10 load increments, where an energy
convergence criterion was used with a 10 –5
convergence tolerance.
After performing the first nonlinear
pushover analysis, the areas where the
pile-soil interface was found to be under
tension were located. Thereafter, the soil
hexahedral elements at those regions were
deleted in order to detach the concrete
from the soil domain. This second model
was developed to further investigate the
mechanical response of the specimens,
where a parametric investigation of the
numerical model was also performed. The
initial model is referred to as the “No Gap”
model, and the second model as the “Gap”
model, as shown in Figure 4. It is important
to note here that the concrete-soil interface
that was found to be under tension extend-
ed to a depth of 3 m (i.e. 5 pile diameters).
MATERIAL PROPERTIES
The material properties adopted for the
soil, concrete and reinforcement numerical
models are presented in this section. It is
important to note that the representative
values of the Young’s modulus and the
compressive strength of the soil in the
numerical analysis were not available from
experimental tests; therefore, a parametric
investigation was performed based on the
values presented in Table 1. As can be
seen, the values assume a minimum soil
compressive strength of 89 kPa, whereas the
maximum assumed uniaxial compressive
strength is 150 kPa. To be able to investigate
the overall response and the effect that the
soil compressive strength has on the SSI
system, five additional values were assumed.
31
Table 2 C
 oncrete (grey section) and steel
(green section) material input 800
parameters as determined from
700
standard material testing
600
Parameter Value Units

Horizontal load (kN)

Young’s modulus 31.4 GPa 500

Uniaxial compressive 400

45 MPa
strength
300
Tensional strength 2 MPa
200
Poisson’s ratio 0.2 –

β for remaining shear 100 Experimental

5 % E = 100 MPa, CS = 110 kPa
strength
0
Elastic limit 50 % 0 20 40 60 80 100 120 140 160
Horizontal displacement (mm)
Young’s modulus 200 GPa

Poisson’s ratio 0.3 – Figure 5 “ Gap” model with the best fit – comparison between the numerical and experimental curves

Steel hardening 4 GPa

Furthermore, three different unsaturated the concrete and steel rebar are presented in
Steel f y 500 MPa soil Young moduli values were investigated Table 2. It is important to note here that for
for the needs of this research work. The both longitudinal and shear reinforcement,
Steel failure 0.12 –
respective material properties defined for the yielding stress was assumed to be equal

Load (kN) Mesh deformation Solid Von Mises strains Load (kN) Mesh deformation Solid Von Mises strains

0.0005
0.000469
0.000438
0.0005
0.000406
0.000469
0.000375
0.000438
0.000344
0.000406
0.000313
0.000375
0.000281
0.000344
100 0.00025
0.000313
0.000219
0.000281
0.000188
0.000156 500 0.00025
0.000125 0.000219
0.0000938 0.000188
0.0000625 0.000156
0.0000313 0.000125
0. 0.0000938
0.0000625
0.0000313
0.
0.0005
0.000469
0.000438
0.000406
0.000375
0.000344
0.000313
0.000281
200 0.00025 0.0005
0.000469
0.000219
0.000188 0.000438
0.000156 0.000406
0.000125 0.000375
0.0000938 0.000344
0.0000625 0.000313
0.0000313 0.000281
0. 600 0.00025
0.000219
0.000188
0.000156
0.0005 0.000125
0.000469 0.0000938
0.000438 0.0000625
0.000406 0.0000313
0.
0.000375
0.000344
0.000313
0.000281
300 0.00025
0.000219
0.000188
0.000156
0.000125
0.0000938 0.0005
0.0000625 0.000469
0.0000313 0.000438
0. 0.000406
700 Plastic hinge
formation
0.000375
0.000344
(load 0.000313
0.000281
0.0005
0.000469 increment 0.00025
0.000219
0.000438
0.000406 prior to 0.000188
0.000375 0.000156
0.000344 failure) 0.000125
0.000313 0.0000938
0.000281 0.0000625
400 0.00025 0.0000313
0.
0.000219
0.000188
0.000156
0.000125
0.0000938
0.0000625
0.0000313
0.
Figure 6 “ Gap” model with the best fit – deformed shapes and Von
Mises strain contours prior to failure

32 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
 600 600

500 500
Horizontal load (kN)

Horizontal load (kN)

400 400

300 300

200 200

100 100

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(a) Horizontal displacement (mm) (d) Horizontal displacement (mm)

Experimental E =50 MPa, CS = 89 kPa E =50 MPa, CS = 100 kPa E =50 MPa, CS = 110 kPa
E =50 MPa, CS = 120 kPa E =50 MPa, CS = 130 kPa E =50 MPa, CS = 140 kPa E =50 MPa, CS = 150 kPa

600 600

500 500
Horizontal load (kN)

Horizontal load (kN)

400 400

300 300

200 200

100 100

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(b) Horizontal displacement (mm) (e) Horizontal displacement (mm)

Experimental E =100 MPa, CS = 89 kPa E =100 MPa, CS = 100 kPa E =100 MPa, CS = 110 kPa
E =100 MPa, CS = 120 kPa E =100 MPa, CS = 130 kPa E =100 MPa, CS = 140 kPa E =100 MPa, CS = 150 kPa

600 600

500 500
Horizontal load (kN)

Horizontal load (kN)

400 400

300 300

200 200

100 100

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(c) Horizontal displacement (mm) (f) Horizontal displacement (mm)

Experimental E =150 MPa, CS = 89 kPa E =150 MPa, CS = 100 kPa E =150 MPa, CS = 110 kPa
E =150 MPa, CS = 120 kPa E =150 MPa, CS = 130 kPa E =150 MPa, CS = 140 kPa E =150 MPa, CS = 150 kPa

Figure 7 H
 orizontal load versus horizontal displacement for different soil material combinations: (a–c) “No Gap” and (d–f) “Gap” models

to 500 MPa, an average value derived from
experimental tests.
NUMERICAL INVESTIGATION
For the purpose of this study, static non-
linear analyses were performed by applying
a lateral load until failure. The numerical
results obtained were compared to the data
obtained from the field experimentation
program. Both the “No Gap” and “Gap”
models were analysed under the same
loading conditions, while the nonlinear
analysis algorithm and the corresponding
parameters were kept the same.
During the physical experiment, the
maximum imposed load was 500 kN,
which was the load level at which the
hydraulic jack reached its maximum exten-
sion; therefore, the experiment had to be
terminated. Consequently, neither the RC
pile nor the soil reached a full failure state,
even though the deformation was larger
than the allowable limit according to inter-
national design codes. For this reason, the
numerical models engaged in this research
study were also used to predict the physi-
cal failure mode of the specimen and the
ultimate capacity that the embedded RC
pile could develop. From the numerical

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 33
results derived using the two numerical
models (“No Gap” and “Gap”), it was found
that the latter resulted in the best match to
the experimental response, i.e. where the
numerically derived lateral load-displace-
ment curve best matched the experimental
data (Figure 5 on p 32). The “Gap” model Z

adopted all the experimental material

Y
parameters obtained from the experiments
performed on the concrete and steel mate- X
rial, while a soil Young’s modulus (E) of
100 MPa and a corresponding compressive
strength (CS) of 110 kPa were assumed.
Therefore, by reverse engineering the
problem, representative soil properties were
determined. Their magnitudes were found
to be realistic for the in-situ soil profile. Figure 8 E lement used for the soil (left) and concrete (right) stress-strain curves
Figure 6 (on p 32) shows the deformed
shapes and the Von Mises strain contours
throughout the loading phase. Specific 180
attention is given to the last two load incre- 160
ments resulting from the nonlinear ulti-
mate limit state analysis of the optimum 140

“Gap” model. It can be seen that increasing 120

deformation developed in the soil domain
Stress (kPa)

in front of the RC pile specimen, indicating
a 3D compressive failure of the clay mate-
rial due to excessive compressive stresses.
This led to large horizontal deformations
that caused the failure of the RC pile’s lon-
gitudinal reinforcement, which was located
in the tensile region of the circular section.
As mentioned above, the physical experi-
ment was terminated at a total horizontal
applied load of 500 kN. The numerical
model predicted failure to occur at a hori-
zontally applied load of 700 kN, thus, the
model allowed a failure mechanism to be
observed.
In order to further investigate the
numerical response of the models devel-
oped, a sensitivity analysis was performed
where the Young’s modulus (ES) for the
soil material was kept constant at values
of 50 MPa, 100 MPa and 150 MPa respec-
tively, while the compressive strength
(CS) of the soil was varied from 89 kPa to
150 kPa (see Table 1). In Figure 7 the total
imposed horizontal load versus the hori-
zontal displacement of the pile can be seen
as derived from the nonlinear analyses. A
comparison with the experimental curve is
also depicted.
As can be seen from Figure 7 (on
p 33), the stiffness of the pile and the soil
structural system decreases with increasing
horizontal load, which is attributed to the
damage development in the two domains.
In addition, an increase in compressive
strength of the soil material results in a
100
80
60
40
20
0
0 20 40
E =100 MPa, CS = 89 kPa
E =100 MPa, CS = 100 kPa
E =100 MPa, CS = 110 kPa
Figure 9 S tress-strain curves of the soil element in front of the pile
stiffer response and hence less horizontal
deformation. It can also be observed that
the “No Gap” model gives a stiffer overall
response attributed to the introduction of
additional stiffness at the interface of the
RC pile and the soil domain. Finally, it is
evident that ReconanFEA (2020) was able
to successfully run all of the developed
models without numerical instabilities;
therefore providing the required insight in
relation to the local and global nonlinear
mechanical behaviour of the RC pile and
the soil domain.
Soil and concrete stress and strain
curves for the most critically loaded ele-
ments in the model were developed to get
a better understanding of the material
response during load application. Figure 8
indicates the elements where the solid Von
Mises stress-strain plots were generated
60 80 100 120 140 160 180 200
Strain 10 –6 (microstrain)
E =100 MPa, CS = 120 kPa E =100 MPa, CS = 150 kPa
E =100 MPa, CS = 130 kPa
E =100 MPa, CS = 140 kPa
(yellow elements for the soil and red for the
pile). These elements are located in front
of the pile and represent the most critically
loaded elements in the numerical analysis.
Figures 9 and 10 represent the stress-strain
plots for the respective soil and concrete
elements determined from the nonlin-
ear analysis. For these plots, the chosen
combination of a soil Young’s modulus of
100 MPa and a compressive strength rang-
ing from 89 kPa to 150 kPa was used for
the soil.
Figure 9 represents the stress-strain
curve for the soil element in front of the
RC pile. It can be observed that each curve
reaches its compressive strength value at a
different strain. The slope (E) of each curve
is relatively similar, whereafter plastic
deformation occurs once each material has
reached its ultimate compressive strength.

34 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
 Since neither strain hardening nor strain
10 000 softening occurs within the soil element,
9 000 the overall mechanical response of the soil
material is classified as an elastic-perfectly
8 000
plastic material.
7 000 Figure 10 represents the stress-strain
curves of the concrete element in front of
6 000
Stress (kPa)

the RC pile. It can be observed that the

5 000 slope in the elastic region (E) is the same
4 000 for the different compressive strength
values. After the concrete has undergone
3 000
elastic deformation, yielding of the con-
2 000 crete occurs at a specific stress, whereafter
1 000 a certain magnitude of strain hardening
occurs until failure. When looking at the
0
0 100 200 300 400 500 600
stress magnitude in the concrete pile at
Strain 10 –6 (microstrain) failure, it can be observed that, for a lower
E =100 MPa, CS = 89 kPa E =100 MPa, CS = 120 kPa E =100 MPa, CS = 150 kPa compressive strength of the soil, a lower
E =100 MPa, CS = 100 kPa E =100 MPa, CS = 130 kPa solid Von Mises stress develops within the
E =100 MPa, CS = 110 kPa E =100 MPa, CS = 140 kPa concrete, while, for a higher compressive
strength of the soil, a higher solid Von
Figure 10 S tress-strain curves of the concrete element in front of the pile Mises stress develops within the concrete
until significant deformation starts to
occur. This phenomenon is attributed to
40 the SSI effect, as well as to the stiffness that
the soil exhibits, which is controlled by the
uniaxial compressive soil strength, which
in turn determines the stress level that
30 develops within the soil domain in front of
the RC pile.
Difference in terms of horizontal displacement (%)

20 PARAMETRIC INVESTIGATION
A sensitivity analysis was conducted on
the parameters which could not be explic-
itly measured in the physical experiment.
10 These parameters were the Young’s modu-
lus (E) and the compressive strength (CS)
of the soil. For this sensitivity analysis,
the “Gap” model with the same geometry,
0
materials, constraints, loading conditions
and nonlinear analysis was used. The best
fit between the numerical result and the
field observation resulted from using an
E value of 100 MPa and a compressive
–10
strength of 110 kPa. These values were
adopted as the base combination for the
sensitivity analysis. This combination
was used to determine how sensitive the
–20
critical parameters were to the horizontal
displacement of the model for different
load increments, which would also provide
insight on how these parameters affect
–30
–0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4
the overall mechanical behaviour of
Change in compressive strength (CS) the model.
100 kN (Change CS) 400 kN (Change CS) 700 kN (Change CS) In order to illustrate the sensitivity of
200 kN (Change CS) 500 kN (Change CS) these parameters, the Young’s modulus
300 kN (Change CS) 600 kN (Change CS) was kept constant, while the compressive
strength was increased and decreased
Figure 11 S ensitivity of the model due to soil compressive strength in terms of horizontal displacement by 30% accordingly. Figure 11 shows the

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 35
sensitivity of the model in terms of com-
pressive soil strength and how the lateral
displacement of the pile varied under dif-
ferent lateral load increments, ranging from
100 kN to 700 kN. Next, the compressive
strength of the soil was kept constant,
while the Young’s modulus was assigned
different values as shown in Figure 12.
Figure 12 represents the sensitivity of the
model in terms of horizontal displacement
under different load increments, again
ranging from 100 kN to 700 kN.
As can be observed in Figure 11, the
higher the load, the higher the horizontal
displacement difference is, compared to
the initial result (best-fit curve). Therefore,
as the load increases, the sensitivity of
the model related to the compressive
strength increases as well. In addition, a
lower compressive strength in the soil is
more sensitive than a higher compressive
strength. Therefore, a higher compressive
soil strength does not affect the model’s
horizontal displacement as much as a
lower soil compressive strength, resulting
in a model that is more sensitive for lower
values. Furthermore, from Figure 12 it can
be observed that a lower E value derives
slightly higher differences to the model
than a higher E value. It can also be seen
that for any E value, the model is affected
more for lower load values (100 kN).
However, for a lower E value, as the load
increases, the E value increase affects the
model’s response more for higher horizon-
tal loads. While for a higher E value, as the
load increases, the E value becomes less
crucial in terms of the numerically derived
horizontal displacement.
A comparison between the sensitiv-
ity of the Young’s modulus (E) and the
compressive strength (CS) of the soil for
each load increment is given in Figure 13.
It can be observed that, for an increase of
the understudied parameters and at a very
low load (100 kN), the Young’s modulus
is more crucial and affects the computed
horizontal displacement more, compared
to the compressive strength of the soil.
However, for the remaining load incre-
ments, the soil’s compressive strength
is the parameter that plays the most
important role in controlling the RC pile’s
deformation that numerically derives from
the model. Therefore, the parametric inves-
tigation reveals that the overall nonlinear
response of the soil domain is found to be
significantly affected by both parameters,
with the soil’s compressive strength being
the most important one.
36
25
20
Difference of the horizontal displacement (%)
15
10
5
0
–5
–10
–15
–0.4 –0.3
100 kN (Change E)
200 kN (Change E)
300 kN (Change E)
Figure 12 S ensitivity of the model due to Young’s modulus in terms of horizontal displacement
CONCLUSIONS
Advanced numerical modelling and
analyses were used to investigate the
nonlinear mechanical behaviour of an RC
pile embedded in unsaturated soil, and sub-
jected to lateral load tests in South Africa.
Two different finite element models were
developed in which the material properties
used in the models were based on material
testing prior to the testing of the RC pile.
The “Gap” model was found to provide the
best fit when compared to the experimen-
tally obtained horizontal load-displacement
curve. Even though the physical experi-
ment was not able to be loaded to failure
due to equipment limitations, the numeri-
cal investigation predicted that the failure
occurred due to a 3D compressive failure of
the soil in front of the RC pile, which led to
a bending failure of the RC pile due to large
horizontal deformations.
A parametric investigation further
showed that the developed numerical
models were affected by the compressive
strength of the soil defined within the
numerical material models. It was also
found that ReconanFEA, based on soil
material reverse-engineered from the prob-
lem at hand, was able to successfully model
the ultimate failure mechanism, allowing
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
–0.2 –0.1 0 0.1 0.2 0.3 0.4
Change in Young's modulus (E)
400 kN (Change E) 700 kN (Change E)
500 kN (Change E)
600 kN (Change E)
the actual soil properties leading to the
overall mechanical response of the SSI
system to be back-calculated. Finally, the
software was found to be able to numeri-
cally handle the numerous nonlinear
analyses that were performed without any
numerical instabilities.
Future work will foresee the construc-
tion of different models, which will pro-
duce numerical results that will be used for
the development of a dataset. This dataset
will be used to train artificial intelligence
and machine learning algorithms for the
development of predictive models that will
be able to estimate the expected stress and
displacement levels for different RC pile
geometries and soil domains.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the UK
Engineering and Physical Sciences Research
Council (EPSRC) Global Challenges Fund
under the Wind-Africa project, Grant
Ref: EP/P029434/1, for the pile load-
displacement data used in this paper. This
work received financial support from the
EuroHPC-JU project EuroCC (G.A. 951732)
of the European Commission. Parts of
the runs were performed on the cyclone
 40

30

20
Change in horizontal displacement (%)

10

–10

–20

–30
–40 –30 –20 –10 0 10 20 30 40
Change in parameters (%)
100 kN (Change E) 500 kN (Change E) 100 kN (Change CS) 500 kN (Change CS)
200 kN (Change E) 600 kN (Change E) 200 kN (Change CS) 600 kN (Change CS)
300 kN (Change E) 700 kN (Change E) 300 kN (Change CS) 700 kN (Change CS)
400 kN (Change E) 400 kN (Change CS)

Figure 13 S ensitivity of the E and CS on the change in horizontal displacement of the model

machine hosted at the HPC National
Competence Center of Cyprus at the
Cyprus Institute (Project pro21b103).
REFERENCES
Atkinson, J 2017. The Mechanics of Soils and
Foundations. Boca Raton, FL: CRC Press.
Breysse, D, Niandou, H, Elachachi, S & Houy, L 2005.
A generic approach to soil-structure interaction
considering the effects of soil heterogeneity.
Geotechnique, 55(2): 143–150.
Červenka, J & Papanikolaou, V K 2008. Three-
dimensional combined fracture-plastic material
model for concrete. International Journal of
Plasticity, 24(12): 2192–2220.
Gravett, D Z & Markou, G 2021. State-of-the-art
investigation of wind turbine structures founded
on soft clay by considering the soil-foundation-
structure interaction phenomenon: Optimization
of battered RC piles. Engineering Structures, 235:
112013.
Gravett, D Z, Mourlas, C, Taljaard, V-L, Bakas, N,
Markou, G & Papadrakakis, M 2021. New
fundamental period formulae for soil-reinforced
concrete structures interaction using machine
learning algorithms and ANNs. Soil Dynamics and
Earthquake Engineering, 144: 106656.
Jardine, R, Potts, D, Fourie, A & Burland, J 1986.
Studies of the influence of non-linear stress-
strain characteristics in soil-structure interaction.
Geotechnique, 36(3): 377–396.
Karthigeyan, S, Ramakrishna, V & Rajagopal, K 2006.
Influence of vertical load on the lateral response of piles
in sand. Computers and Geotechnics, 33(2): 121–131.
Kausel, E 2010. Early history of soil-structure
interaction. Soil Dynamics and Earthquake
Engineering, 30(9): 822–832.
Kavitha, P, Beena, K & Narayanan, K 2016. A review on
soil-structure interaction analysis of laterally loaded
piles. Innovative Infrastructure Solutions, 1(1): 1–15.
Kotsovos, M D & Pavlovic, M N 1995. Structural
Concrete: Finite-element Analysis for Limit-state
Design. London: Thomas Telford.
Markou, G, Alhamaydeh, M & Saadi, D 2018. Effects
of the soil-structure-interaction phenomenon on
RC structures with pile foundations. Proceedings,
9th Greek Association of Computational Mechanics

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 37
 (GRACM) International Congress on Computational
Mechanics. 4–6 June 2018, Chania, Greece.
Markou, G & Papadrakakis, M 2013. Computationally
efficient 3D finite element modeling of RC structures.
Computers and Concrete, 12(4): 443–498.
Markou, G & Roeloffze, W 2021. Finite element
modelling of plain and reinforced concrete
specimens with the Kotsovos and Pavlovic material
model, smeared crack approach and fine meshes.
International Journal of Damage Mechanics, 30(6):
845–871. doi: 10.1177/1056789520986601.
38
Menegotto, M 1973. Method of analysis for cyclically
loaded RC plane frames including changes in
geometry and non-elastic behavior of elements
under combined normal force and bending.
Proceedings, International Association for Bridge
and Structural Engineering (IABSE) Symposium on
Resistance and Ultimate Deformability of Structures
Acted on by Well Defined Repeated Loads. Lisbon,
Portugal, pp 15–22.
Oanta, E M, Panait, C & Raicu, A 2016. Original data
preprocessor for Femap/Nastran. In Vladescu,
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
M, Panait, C T, Tamas, R, Caruntu, G & Cristea,
I (Eds). Advanced Topics in Optoelectronics,
Microelectronics, and Nanotechnologies
(Conference). Vol 10010. 5–28 August 2016,
Constanta, Romania. Bellingham, WA: SPIE
Publications, p 1001020.
ReconanFEA v2.00. User’s Manual, 2020.
Available at: https://www.researchgate.net/
publication/342361609_ReConAn_v200_Finite_
Element_Analysis_Software_User%27s_Manual.
 TECHNICAL NOTE
Comparison of sinkhole
size versus borehole
JOURNAL OF THE SOUTH AFRICAN
INSTITUTION OF CIVIL ENGINEERING

hazard classification
ISSN 1021-2019 (print) | ISSN 2309-8775 (online)
Vol 65 No 2, June 2023, Pages 39–43, Note 1578

JANET BUNK completed a BSc (Hons) in

J Bunk Geology and Engineering Geology in 1999.
She started her career in the mining
environment with SRK and was seconded to
Navachab, Namibia, to monitor the stability
Sinkholes are a common occurrence on dolomitic land in South Africa. Different methods have of the mine until 2002. She then did
contract work for various companies,
been used in the past to classify the risk of sinkhole occurrence. The inherent hazard classifications
amongst others Aurecon/Zutari, until 2018
(IHC) of boreholes are compared in this article to actual sinkhole occurrences. The national when she joined the Council for Geoscience. She is currently involved with
sinkhole database and the ENGEODE database of boreholes, both belonging to the Council for research on dolomite, ranging from zonation efforts from available data,
Geoscience (CGS), were used for this comparison between the IHC of the borehole and the size of fundamental testing of materials, early warning systems for sinkholes,
advising and monitoring of sinkholes in the country, and shallow
the sinkhole that occurred within a 50 m radius. Thus, the sinkholes in the database were used for
geotechnical investigations.
this comparison that had a borehole located within 50 m from the sinkhole.
Contact details:
Four regions were identified and described separately, namely the Far West Rand, the
Council for Geoscience
Tshwane and Ekurhuleni Municipalities as separate areas, and the fourth group as the Private Bag X112
Johannesburg and Sedibeng Municipalities and the Mpumalanga, North West and Northern Pretoria 0001
Cape Provinces. Although a very high percentage has been underestimated (ranging from 42% South Africa
E: jbunk@geoscience.org.za
to 59% for the different regions), there is a relatively good IHC correlation between the hazard
classification of the boreholes and the sinkholes (between 56% to 67%). Another problem is that
an IHC of 1 to 4 (low to medium risk) was assigned to the boreholes in 17% to 29% of the cases
for the different areas, yet sinkholes still formed. These areas are more problematic, since the
risk from the drilling has been underestimated.
The current methodology to classify boreholes has not proved to be one hundred percent
accurate to predict the size of the sinkhole for the area, and several factors have been highlighted
that should be taken into account when boreholes are classified. Further research needs be
carried out around the discrepancies between the IHC of boreholes compared to actual sinkholes.

INTRODUCTION Schalkwyk (1995) and Buttrick et al (2001).

Sinkholes are a common occurrence on These have been summarised in Part 2 of
dolomitic land in South Africa. Dolomite the SANS 1936-2 (SANS 2012).
is a rock that has been subjected to long The inherent hazard classes are deter-
periods of chemical weathering (Sartain mined with respect to the inherent sus-
et al 2011). This has resulted in cavities in ceptibility (from low to high) of material
the dolomite and the formation of highly to mobilise, and subsequent formation of a
variable residuum. Wad is a dolomitic specific size sinkhole or subsidence (small
residuum which is weak in certain areas, to very large). This is based on the calculated
highly erodible and compressible, and has potential size of these surface instability
a low density. When the residuum consists features and their inherent susceptibility to
of weak wad (Wagener 1982) or cavities are mobilise as determined for individual bore-
present, the profile has a higher suscepti- hole profiles in SANS 1936-2 (SANS 2012).
bility to sinkhole formation. The IHC for sinkhole and subsidence
In order to mitigate the risks associ- development can be expressed in terms of
ated with sinkholes, boreholes are drilled eight categories (IHC 1 to 8), which denote
and the soil/rock profile encountered is the probability of a sinkhole or subsidence
assessed and classified in terms of the of a specific size occurring, as summarised
inherent hazard for sinkhole formation. in Table 1. These eight classes are grouped
Different methods were used in the past to into three broad categories, namely low
classify this risk, such as the “Method of (IHC 1), medium (IHC 2 to 4) and high
Scenario Supposition” by Buttrick (1992) (IHC 5 to 8) according to SANS 1936-2
with follow-up papers by Buttrick and Van (SANS 2012).

Bunk J. Comparison of sinkhole size versus borehole hazard classification. Keywords: dolomite, sinkhole, hazard comparison, borehole
J. S. Afr. Inst. Civ. Eng. 2023:65(2), Art. #1578, 5 pages. http://dx.doi.org/10.17159/2309-8775/2023/v65n2a5

The Journal of the South African Institution of Civil Engineering, which is distributed internationally, is a peer-reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND). Readers may therefore freely use and share the content as long as they credit the original creator and publisher, do
not change the material in any way, and do not use it commercially. Copyright of this article remains with the authors. 39
 The manifestation of a sinkhole or sub-
sidence at ground surface does not always
define the depth and lateral extent of an
instability (Kleinhans & Van Rooy 2016).
The receptacle at depth that accommodates
the eroded material may not be directly
below the area of visual impact on the sur-
face, but can be located at distances more
than 20 m from the sinkhole or subsidence
manifestation at surface. The evaluation of
an event should therefore consider covering
a surface area of between 20 m to 50 m
around the sinkhole or subsidence area to
ensure that the area of influence has been
covered (Kleinhans & Van Rooy 2016).
Table 1 Definition of the eight inherent hazard classes (after Buttrick et al 2001)
Inherent
hazard Characterisation of area
class (IHC)
Areas characterised as reflecting a low inherent hazard of sinkhole and subsidence
Class 1
formation (all sizes) with respect to ingress of water
Areas characterised as reflecting up to a medium inherent hazard of small-sized
Class 2
sinkhole (< 2 m) and subsidence formation with respect to ingress of water
Areas characterised as reflecting up to a medium inherent hazard of medium-sized
Class 3
sinkhole (2–5 m) and subsidence formation with respect to ingress of water
Areas characterised as reflecting up to a medium inherent hazard of large-sized
Class 4
sinkhole (5–15 m) and subsidence formation with respect to ingress of water
Areas characterised as reflecting up to a high inherent hazard of small-sized sinkhole
Class 5
(< 2 m) and subsidence (all sizes) formation with respect to ingress of water

Areas characterised as reflecting up to a high inherent hazard of medium-sized

Class 6
sinkhole (2–5 m) and subsidence formation with respect to ingress of water
PROBLEM STATEMENT
Areas characterised as reflecting up to a high inherent hazard of large-sized sinkhole
This paper discusses the efficiency of using Class 7
(5–15 m) and subsidence formation with respect to ingress of water
point data to predict sinkhole hazard and
size. This is used to compare the classified Areas characterised as reflecting up to a high inherent hazard of very large-sized
Class 8
sinkhole (> 15 m) and subsidence formation with respect to ingress of water
hazard of the borehole with the size of the
sinkhole that has occurred within a 50 m
radius from the borehole. The size of the Table 2 Classification of sinkhole size (after Buttrick et al 2001)
sinkhole is defined as the volume of the Maximum diameter of
sinkhole and will be explained later on in Maximum potential Suggested
surface manifestation
development space terminology
this paper. (dimension: metres)

Small potential development space <2 Small sinkhole

Medium potential development space 2–5 Medium-sized sinkhole

METHODOLOGY
The sinkhole data and the borehole logs Large potential development space 5–15 Large sinkhole
from the ENGEODE database have been
Very large potential development space > 15 Very large sinkhole
used for this study. There are currently

18%
Tshwane Ekurhuleni
267 sinkholes 31% 106 sinkholes
42%
Overestimated Overestimated
57% 24% Correct Correct
Underestimated Underestimated
27%

Far West Rand 24% Other provinces

36% 156 sinkholes 17 sinkholes
46%
Overestimated Overestimated
Correct 59% Correct
18%
Underestimated Underestimated
19%

Figure 1 C
 omparison of IHC values of boreholes versus IHC values of actual sinkhole volumes, i.e. borehole IHC is either overestimating or
underestimating actual sinkhole volume

40 Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
 Tshwane, 267 sinkholes Ekurhuleni, 106 sinkholes
70 35
60 30
Number of sinkholes

Number of sinkholes
50 25
40 20
30 15
20 10
10 5
0 0
–3 –2 –1 0 1 2 3 4 5 6 –3 –2 –1 0 1 2 3 4 5 6 7
IHC sinkhole compared to IHC borehole IHC sinkhole compared to IHC borehole

FWR, 156 sinkholes Other provinces, 17 sinkholes

35 6
30
Number of sinkholes

Number of sinkholes
5
25
4
20
3
15
2
10
5 1
0 0
–3 –2 –1 0 1 2 3 4 5 6 7 –2 –1 0 1 2 3
IHC sinkhole compared to IHC borehole IHC sinkhole compared to IHC borehole

Figure 2 D
 ifference between IHC calculated from sinkhole volume and classified from borehole log (thus IHC sinkhole value minus IHC borehole value)

5 171 sinkhole entries in the database, but
only 546 have been used in this study. The
limiting factors were dimensions that were
not available for all entries, and the fact
that a borehole had to be present within
50 m from the sinkhole.
The sinkhole database includes all
sinkholes in South Africa and is subdivided
into different municipalities or regions,
namely the Far West Rand region, Tshwane,
Ekurhuleni, Johannesburg and Sedibeng
Municipalities and the Mpumalanga, North
West and Northern Cape Provinces. Four
regions were analysed: the Far West Rand,
Tshwane and Ekurhuleni Municipalities as
separate areas, while the rest of the data was
combined to form a fourth group. For this
comparison, sinkholes were selected that
were up to 50 m away from a borehole. In
a couple of cases, the hazard classification
of boreholes in the vicinity was used if the
information on the borehole was insuf-
ficient or not representative of the area.
To compare the sinkholes with the
borehole classification, an IHC value was
assigned to each sinkhole, based on the
diameter of the sinkhole and the volume
of the sinkhole. The maximum volume for
a small, medium and large sinkhole was
calculated from the dimensions as specified
by Buttrick et al (2001). These values are
tabulated in Table 2, and a minimum depth
of 1 m was used to calculate the volume
of the size sinkhole. It was found that the
assigned IHC value based on the volume
differed less from the IHC of the borehole
than the assigned IHC based on the diam-
eter. For this study, only the IHC based on
the volume of the sinkhole was taken into
consideration. It was further assumed that
the area around the sinkhole reflected a
high inherent hazard, thus IHC values of 5
to 8 were assigned to the sinkholes.
The comparison of the IHC values was
thus undertaken from two perspectives:
(a) to determine whether the hazard has
been accurately predicted (thus IHC values
of 5 to 8 and not IHC values less than 5),
and (b) whether the size of the sinkhole
corresponds to the IHC value of the bore-
hole. The percentages of the overestimated,
correct and underestimated classifications
of the boreholes versus the sinkholes are
presented in Figure 1.
The classifications of the boreholes
were verified individually. Nine percent of
the classifications were corrected, implying
that a higher hazard was allocated than the
initial classification, while four percent were
borderline IHC 3/4 or IHC 6, in which case
the latter was chosen as the inherent hazard
classification. Twelve percent of the classifi-
cations were correct, but were changed to a
higher hazard classification due to the clas-
sification of the surrounding boreholes; thus
the geological conditions of the boreholes
in the area were worse than the borehole
originally classified. In total, an adjustment
was made to 25% of the classifications of
the boreholes in the database. For the rest
of the classifications, only the borehole log
was taken into account and not the general
geological conditions of the area.
The groundwater table was incorporated
in the classifications as per the method pro-
posed by Oosthuizen and Van Rooy (2015).
The comparison in this paper considered the
ingress scenario of the boreholes and did not
take the dewatering scenario into account.
COMPARISON
Direct correlation of the IHC values
The inherent hazard classifications (IHCs)
of the individual boreholes and the IHCs of
the corresponding sinkholes are compared
in Figure 1. Only approximately one quarter
is deemed classified correctly (23%), while
half of the IHCs of the boreholes are gener-
ally underestimating the size and risk of the
sinkhole, with 26% of the boreholes overesti-
mating the sinkhole size and/or hazard.
Although only a small representation
is classified correctly, 61% of the classifica-
tions in the Tshwane Municipality area dif-
fered by one value, i.e. one class up or down
from the sinkhole hazard classification.
This represents 67% for the Ekurhuleni
Municipality, 56% for the Far West Rand
region and 65% in the other provinces, as
illustrated in Figure 2. This indicates that

Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023 41
Figure 3 V
 isual inherent hazard classification for the Waterkloof Airbase, Tshwane Municipality
there is a relatively good IHC correlation
between the hazard classification of the
boreholes and the sinkhole sizes.
Classifications with too low risk
As discussed previously, the areas where
sinkholes occur should be classified in
terms of high risk. This suggests that IHC
values from 1 to 4 are not deemed satisfac-
tory in an area where sinkholes occur. In
the Tshwane Municipal area, 22% of the
boreholes were classified with IHC values
of 4 and lower, suggesting a low to medium
hazard for sinkhole formation. For the
Ekurhuleni Municipality, this figure is 17%,
in the Far West Rand region 24% and for
the other provinces 29%.
The correlation between the clas-
sification of the borehole and those of the
sinkhole is visually indicated in Figure 3.
The larger circle is colour-coded for the
IHC of the sinkholes, while the smaller
circle represents the IHC of the borehole.
Where the colours are the same, it suggests
a good correlation for hazard classification
between the sinkhole and the borehole in
proximity thereof. However, if a lighter col-
our is present within the sinkhole marking,
42
it suggests that the size of the sinkhole has
been underestimated. The blue, green and
yellow dots indicate that the hazard has
been undervalued as either low or medium
risk for sinkhole formation, thus IHC 1 to 4
classifications.
CONCLUSION
Although a very high percentage has been
underestimated (ranging from 42% to
59% for the different regions), there is a
relatively good IHC correlation between
the hazard classification of the boreholes
and the sinkholes. The number of classifi-
cations that are correct and differ by one
value, is 61% in the Tshwane Municipality
area, 67% for the Ekurhuleni Municipality,
56% for the Far West Rand region and
65% in the other provinces. The current
methodology to classify the boreholes has
not proved to be one hundred percent
accurate to predict the size of the sinkhole
for the area.
The other fact is that an IHC of 1 to 4
(low to medium risk) was assigned to the
boreholes in 17% to 29% of the cases for
the different areas. These areas are more
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
problematic, since the risk has been under-
estimated from the drilling.
The areas where discrepancies occurred
have been assessed in terms of the subsur-
face conditions and hazard classifications
of the boreholes in the immediate vicinity.
During the correction of the classifications
of the boreholes, the following factors were
noted:
QQ Many boreholes were too shallow to
classify appropriately (30 m and less in
depth).
QQ Classification changed when bore-
holes in the area had a higher hazard
classification.
QQ Deep competent overburden profiles,
but underlain by cavities and/or weak
soil.
QQ Deep competent profiles, especially in
the Far West Rand, but still sinkholes
occurred in the area.
QQ Shallow dolomite in an area, but the
hazard not classified as high.
QQ Boreholes classified as borderline
between high and medium hazard.
It is important to take note that when a
profile has cavities or weak soil in the pro-
file, even if a thick overburden is present,
sinkholes can occur. These boreholes
should rather be classified with IHC 6
or higher to incorporate the high risk of
sinkhole occurrence. In a shallow dolomitic
area, the risk for small sinkholes to form is
high, and the classification of the boreholes
should reflect that.
It is important to take cognisance of the
boreholes in the vicinity when assessing
a site. Boreholes with a high hazard can
contribute to the instability of the area. In
an area with shallow dolomite bedrock, the
risk for the occurrence of small sinkholes
is higher than generally anticipated. Deep
competent profiles have a profound influ-
ence on the instability of the area, and
alternative ways to determine the stability of
these areas should be investigated. Another
factor that might be included in this process,
is to determine whether the type of over-
burden has an influence on the formation of
sinkholes and whether it can be recognised
Journal of the South African Institution of Civil Engineering
in the borehole or other laboratory tests.
Further research needs to be carried out
around the discrepancies between the IHC
of boreholes compared to actual sinkholes.
REFERENCES
Buttrick, D B 1992. Characterization and appropriate
development on sites on dolomite. PhD Thesis.
University of Pretoria.
Buttrick, D B & Van Schalkwyk, A 1995. The method
of scenario supposition for stability evaluation of
sites on dolomitic land in South Africa. Journal of
the South African Institution of Civil Engineering,
37(4): 4–14
Buttrick, D B, Van Schalkwyk, A, Kleywegt, R J &
Watermeyer, R B 2001. Proposed method for
dolomite land hazard and risk assessment in South
Africa. Journal of the South African Institution of
Civil Engineering, 43(2): 27–36.
Kleinhans, I & Van Rooy, J L 2016. Guidelines for
sinkhole and subsidence rehabilitation based
Volume 65 Number 2 June 2023
on generic geological models of a dolomite
environment on the East Rand, South Africa.
Journal of African Earth Sciences, 117: 86–101.
Oosthuizen, T & Van Rooy, J L 2015. Hazard of
sinkhole formation in the Centurion CBD using the
simplified method of scenario supposition. Journal
of the South African Institution of Civil Engineering,
7(2): 69–75.
SANS (South African Standard) 2012. SANS.
1936-2: Development of Dolomitic Land. Part 2:
Geotechnical Investigations and Determinations.
Pretoria: SABS Standards Division.
Sartain, N, Mian, J, O’Riordan, N & Storry, R 2011.
Case study on the assessment of sinkhole risk for
the development of infrastructure over karstic
ground. In Vogt, N, Schuppener, B, Straub, D
& Bräu, G (Eds). Proceedings, 3rd International
Symposium on Geotechnical Safety and Risk (ISGSR
2011). 2–3 June 2011, Munich, Germany.
Wagener, F von M 1982. Engineering construction on
dolomites. PhD Thesis. Pietermaritzburg: University
of KwaZulu-Natal.
43
 Guidelines for the preparation of
papers and technical notes
INTRODUCTION
The Journal of the South African Institution of Civil Engineering,
which is a peer-reviewed, open-access journal, is published quar-
terly in March, June, September and December. Articles submitted
for publication are reviewed by a panel of referees under the guid-
ance of the SAICE Journal Editorial Panel. The journal publishes
research papers covering all the disciplines of civil engineering
(structural, geotechnical, railway, coastal/marine, water, construc-
tion, environmental, municipal, transportation) and associated
topics that are relevant to the civil engineering profession, and
that preferably have particular relevance to civil engineering in
Southern Africa and the African continent. When preparing arti-
cles for publication, authors should please take note of, and comply
with, the guidelines as set out below.
CLASSIFICATION OF ARTICLES
CONSIDERED FOR PUBLICATION
QQ Technical papers are well-researched, in-depth, fully refer-
enced technical articles not exceeding 6 000 words in length
(excluding tables, illustrations and the list of references).
Related papers that deal with ‘softer sciences’ (e.g. education,
social upliftment, etc) are accepted if they are of a techni-
cal nature and of particular interest to the civil engineering
profession. The latter type of paper will be subject not only
to peer-review by civil engineers, but also to review by non-
engineering specialists in the field covered by the paper.
QQ Technical notes are short, fully referenced technical articles
that do not exceed 2 000 words. A typical technical note will
have limited scope often dealing with a single technical issue
of particular importance to civil engineering.
QQ Review papers are considered for publication as either techni-
cal papers or technical notes on condition that they are the
original work of the author and will assist the reader with the
understanding, interpreting, or applying of the subject under
review. A review paper must contain criteria by which the work
under review was evaluated and must contribute by synthesis-
ing the information and drawing new conclusions from the
dissemination of the previously published work.
QQ Discussions on published articles are welcomed up to six
months after publication. The length of discussion contribu-
tions is limited to 1 500 words. Where appropriate, discussion
contributions will be subject to the normal reviewing process
and will be forwarded to the authors of the original article
for reply.
POLICY REGARDING LANGUAGE AND
ORIGINALITY OF SUBMITTED ARTICLES
QQ Language: Manuscripts should be presented in English, as the
journal is distributed internationally.
QQ Original work: Papers and technical notes must be original
contributions. Authors of JSAICE articles must affirm that
44
submitted material has not been published previously, is not
under consideration for publication elsewhere and will not be
submitted elsewhere while under consideration by the SAICE
Journal Editorial Panel. It is the responsibility of the authors
to ensure that publication of any paper in the journal will not
constitute a breach of any agreement or the transgression of
any law. The corresponding author should confirm that all
co-authors have read and approved the manuscript and accept
these conditions. Authors are responsible for obtaining permis-
sion to publish experimental data and other information that
may be confidential or sensitive. Authors are also responsible
for obtaining written permission from copyright owners (usu-
ally the publishers) when reproducing material that had been
published elsewhere. If such material had been published under
a Creative Commons Licence, appropriate recognition must be
included in the submitted manuscript.
SUBMISSION PROCEDURES AND
REQUIRED FORMAT
QQ Online submission: Manuscripts must be uploaded as PDF
files (http://journal.saice.org.za). Individual file sizes may
not exceed 10 MB. Should you experience problems when
uploading your paper, please contact the administrative editor
(verelene@saice.org.za).
QQ Format: Manuscripts should be prepared in MS Word and
presented in double-line spacing, single-column layout with
25 mm wide margins. Line numbers must be applied to the
whole document. All pages should bear the authors’ names and
be numbered at the bottom of the page. With the exception of
tables and figures (see below) the document should be typed in
Times New Roman 12 pt font. Contributions should be accom-
panied by an abstract of not more than 200 words.
QQ First page: The first page of the manuscript should include
the title of the paper, the number of words of the main text
(i.e. excluding figures, tables and the list of references), the
initials, titles and surnames of the authors, professional status
(if applicable), SAICE affiliation (Member, Fellow, Visitor,
etc), ­telephone numbers (landline and mobile), and e-mail
and postal addresses. The name of the corresponding author
should be underlined. Five keywords should be suggested.
QQ Figures, tables, photos and illustrations:
QQ These should preferably be submitted in colour, as the jour-
nal is a full-colour publication.
QQ Their positions should be clearly marked in the text as fol-
lows: [Insert Figure 1]
QQ Figures, tables, photos, illustrations and equations should
be numbered consecutively and should appear in the text
directly after they have been referred to for the first time.
QQ Illustrations must be accompanied by appropriate captions.
Captions for tables should appear above the table. All other
captions should appear below the illustration (figures,
graphs, photos).
Volume 65 Number 2 June 2023 Journal of the South African Institution of Civil Engineering
 QQ Only those figures and photographs essential to the under-
standing of the text should be included. All illustrations
should be referred to in the text.
QQ Figures should be produced using computer graphics. Hand-
drafted figures will not be accepted. Lettering on figures
should be equivalent to a Times New Roman 9 pt font, or
slightly larger (up to 12 pt) if desired. Lettering smaller than
9 pt is not acceptable.
QQ Tables should be typed in Times New Roman 9 pt font. They
should not duplicate information already given in the text,
nor contain material that would be better presented graphi-
cally. Tabular matter should be as simple as possible, with
brief column headings and a minimum number of columns.
QQ Mathematical expressions and presentation of symbols:
QQ Equations should be presented in a clear form which can
easily be read by non-mathematicians. Each equation
should appear on a separate line and should be numbered
consecutively.
QQ Symbols should preferably reflect those used in Microsoft
Word Equation Editor or Mathtype, or should be typed using
the Times New Roman symbol set.
QQ Variables in equations (x, y, z, etc, as well as lower-case Greek
letters) should be presented in italics. Numbers (digits),
upper-case Greek letters, symbols of metric measurement
units (m for metres, s for seconds, etc) and mathematical/
trigonometrical functions (such as sin, cos and tan) are not
written in italics, but in upright type (Roman). Variables
and symbols used in the body of the text should match the
format used in the equations, i.e. upright or italics, whichever
is applicable.
QQ Metric measurement abbreviations/units should conform to
international usage – the SI system of units should be used.
QQ Decimal commas may be used, but decimal points are pre-
ferred in accordance with accepted international usage.
QQ Symbols should preferably be defined in the text, but if this is
not feasible, a list of notations may be provided for inclusion
at the end of the paper.
QQ Headings: Sections and paragraphs should not be numbered.
The following hierarchy of headings should be followed:
HEADING OF MAIN SECTION
Heading of subsection
Heading of sub-subsection
QQ References: References should follow the Harvard system. The
format of text citations should be as follows: “Jones (1999) dis-
covered that …” or “Recent results (Brown & Carter 1985; Green
et al 1999) indicated that …”
 References cited in the text should be listed in alphabetical
order at the end of the paper. References by the same author
Journal of the South African Institution of Civil Engineering Volume 65 Number 2 June 2023
should be in chronological order. The following are examples of
a journal article, a book and a conference paper:
QQ Donald, A W 1954. Pore pressure coefficients.
Géotechnique, 4(4): 143–147.
QQ Terzaghi, K 1943. Practical soil mechanics. New York: Wiley.
QQ Smith, A W, Black, C & White, A 1999. Factors affecting the
strength of soils. Proceedings, ASCE Conference on Shear
Strength of Soils, Colorado, pp 503–532.
Papers published previously in the Journal of the South African
Institution of Civil Engineering should be cited if applicable.
QQ Footnotes, trade names, acronyms, abbreviations: These
should be avoided. If acronyms are used, they should be defined
when they first appear in the text. Do not use full stops after
abbreviations or acronyms.
QQ Return of amended papers: Papers requiring amendments will
be accepted up to six months after the referee reports had been
sent to authors, after which time the paper will be withdrawn
from the system.
FINAL ARTICLE
QQ Proofs: First proofs of accepted papers will be sent to authors
in PDF format for verification before publication. No major
re-writes will be allowed at this stage, only essential minor
corrections.
QQ Photos of authors: The final corrected version of the paper
should be accompanied by recent, high-resolution head and
shoulders colour photographs and a profile not exceeding 100
words for each of the authors.
QQ Copyright: Copyright remains with the authors. As a peer-
reviewed, open-access journal licensed under a Creative
Commons Attribution Licence (CC BY-NC-ND) (https://
creativecommons.org/licenses/by-nc-nd/4.0/) the content
of this journal is available to users free of charge. Users are
therefore allowed to copy and distribute the material in any
medium or format, but such material may only be used in
unadapted form and for non-commercial purposes, and due
recognition must be given to the authors, as well as to the
Journal of the South African Institution of Civil Engineering as
the original publisher.
QQ Repository policy (depositing of articles by authors after
publication): After publication a PDF of their published article
is sent to authors for their records. Authors may then upload the
published article to their work/institution website, or any other
repository of their choice, such as, for example, ResearchGate
and/or similar. Only the published version of the paper may be
used/uploaded in this manner, i.e. not the originally submitted
or the accepted/pre-published version.
45