Volume 64 Number 1

March 2022

QQ  mathematical model for determining engineering

A
soil classifications from pedological data
QQ Utilising Land Type data for geotechnical investigations
QQ I ncorporation of additional information into
the South African Wind Load Formulation
QQ  ssessment of local sewage sludge ash as a
A
supplementary cementitious material – effects of
incineration temperature and cooling rate of the ash
QQ E valuation of the seismic response of a reinforced concrete
footing with stub column to increasing peak ground
acceleration using pseudo-dynamic experimentation
QQ  hat leads to severe multi-vehicle crashes on
W
mountainous expressways in Western China?
Volume 64 No 1 March 2022 ISSN 1021-2019

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
Adjunct Professor of Geotechnical Engineering
University of Stellenbosch
Tel +27 11 519 0200
day@jaws.co.za
2 A mathematical model for determining engineering
joint EditorS-in-chief
Prof Chris Clayton soil classifications from pedological data
University of Southampton
c.clayton@soton.ac.uk G C Fanourakis
Prof Gerhard Heymann
University of Pretoria
gerhard.heymann@up.ac.za 12 Utilising Land Type data for geotechnical
Administrative Editor
Verelene de Koker investigations
Tel +27 11 805 5947, Mobile +27 83 378 3996
verelene@saice.org.za G C Fanourakis
journal editorial Panel
Prof Y Ballim – University of the Witwatersrand
Prof W Burdzik – Retired from the University of Pretoria 25 Incorporation of additional information into
Prof CRI Clayton – University of Southampton
Dr P Day – Jones & Wagener (Pty) Ltd (chairman)
the South African Wind Load Formulation
Prof J du Plessis – University of Stellenbosch
Prof GC Fanourakis – University of Johannesburg
F P Bakker, N de Koker, C Viljoen
Prof PJ Gräbe – University of Pretoria
Dr C Herold – Umfula Wempilo Consulting
Prof G Heymann – University of Pretoria 37 Assessment of local sewage sludge ash as a
Prof A Ilemobade – University of the Witwatersrand
Prof SW Jacobsz – University of Pretoria supplementary cementitious material – effects of
Prof EP Kearsley – University of Pretoria
Prof CP Roth – University of Pretoria incineration temperature and cooling rate of the ash
Prof W Steyn – University of Pretoria
Mr M van Dijk – University of Pretoria R Juala, Y Ballim, J Mulopo
Prof C Venter – University of Pretoria
Prof A Visser – University of Pretoria
Dr E Vorster – Zutari
Prof J Wium – University of Stellenbosch
48 Evaluation of the seismic response of a
Prof A Zingoni – University of Cape Town reinforced concrete footing with stub column
Prof M Zuidgeest – University of Cape Town
Peer reviewing
to increasing peak ground acceleration
The Journal of the South African Institution of
Civil Engineering is a peer-reviewed journal
using pseudo-dynamic experimentation
that is distributed internationally S M Hossell, C P Roth
Design and reproduction
Marketing Support Services, Ashlea Gardens, Pretoria
Papers for consideration should be submitted 63 What leads to severe multi-vehicle crashes on
online at: http://journal.saice.org.za
mountainous expressways in Western China?
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed Y Wang, L Wang, L Sun
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 1

TECHNICAL PAPER A mathematical model for
Journal of the South African
Institution of Civil Engineering determining engineering
ISSN 1021-2019
Vol 64 No 1, March 2022, Pages 2–11, Paper 1286 soil classifications from
PROF GEORGE FANOURAKIS (FICE, FSAICE) is
Associate Professor in the Department of
pedological data
Civil Engineering Technology at the (now)
University of Johannesburg, which he joined
around 28 years ago after leaving the G C Fanourakis
employ of Jones and Wagener (Pty) Ltd
Consulting Engineers. He holds the degrees
of MSc(Eng) and DTech(Eng), and is a
Chartered Civil Engineer and Fellow of the Institution of Civil Engineers Various pedological soil classification systems exist worldwide, including an internationally
(UK). He is also a Fellow of the South African Institution of Civil accepted system and various national systems, many of which have been incorporated into
Engineering (SAICE), an Honorary Fellow (and Past-President) of the databases that include maps. Various physical and chemical soil properties are used for
Institute of Professional Engineering Technologists (IPET) and Member of
classifying soils according to these pedological systems. This paper proposes an approach
the Soil Science Society of Southern Africa. His professional participation
further includes membership of three South African Bureau of Standards which may be used to determine the engineering properties of soils from the physical and
(SABS) committees, membership of COM 9: Dissemination of Knowledge of chemical properties that are used to pedologically classify soils by systems, and, in particular,
the fib (fédération internationale du béton), and membership of TC:306 the South African Binomial System. These engineering properties include the USCS and
(Geo-Engineering Education) of the International Society of Soil Mechanics
and Geotechnical Engineering (ISSMGE).
AASHTO classification groups which may, in turn, be used as a means of rapidly determining the
general suitability of areas for proposed development, particularly during the reconnaissance
Contact details:
Department of Civil Engineering Technology
investigation stages of transportation route locations and township developments, with a
University of Johannesburg resultant saving of time and money. The model was verified using data from the study area, as
PO Box 17011 well as from an area located approximately 190 km from the study area. A total of 88% of the
Doornfontein
classification groups determined by the model, in the study area, were correct. Furthermore,
2028
South Africa only 6% of the classifications were incorrect by a maximum of two groups. The classifications
T: +27 11 559 6416 determined for the soils outside the study area were all correct.
E: georgef@uj.ac.za

INTRODUCTION been developed from 1977 to 2018

Pedology and pedological
classification systems
Pedology is a branch of soil science
which deals with the scientific study of
soils, without a view to any particular
application.
A number of pedological soil classifica-
tion systems have been established globally,
which are used as a basis for soil maps.
These include an international system
and many national systems, as detailed by
Fanourakis (2012).
The international system is the World
Reference Base for Soil Classification
(IUSS-WRB 1998; 2006; 2007; 2015). This
system has been translated into a number
of languages, including Arabic, German,
Polish, Russian, Slovak, Spanish and
Turkish (IUSS-WRB 2015). This system
is available in Portable Document Format
(pdf), as well as a Mobile Application (App)
for Android, Apple and Windows mobile
in English for certain countries or regions
(FAO 2021). This App is currently not
available for South Africa.
Keywords: pedology, soil classification, geotechnical engineering, Three versions of the South African
models pedological classification system have
(Fanourakis 2012).
The commonly used South African
Classification System is the older version,
titled “Soil Classification: A Binomial
System for South Africa” (MacVicar et al
1977). An overview of this classification is
given by Fanourakis (2012). This system
has been used as a basis for soil mapping in
South Africa.
According to Calitz (2021), the older
Binomial Classification System (after
MacVicar et al 1977) and the latest Natural
and Anthropogenic System (SCWG 2018)
are the most useful with regard to the
nature and behaviour of soils for geotech-
nical purposes.
Pedological databases
Various international, regional and national
databases comprising pedological, climatic,
topographical and other information have
been developed, as detailed by Fanourakis
(2012). The South African database, which
is known as the Land Type Survey, was
compiled by the Institute of Soil Climate
and Water (LTSS 2021).
The soil data included in the Land Type
Survey of South Africa (LTSS 2021) shows

Fanourakis GC. A mathematical model for determining engineering soil classifications from pedological data.
2 J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #1286, 10 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64n1a1
maps based on the Binomial Classification
System (after MacVicar et al 1977) and
accompanying memoirs which contain,
in decreasing detail, information on the
soils, terrain, geology and climate. The
soil information comprises soil profiles, as
well as quantitative physical and chemical
information of the soils with their repre-
sentative profiles (termed modal profiles).
Comprehensive details on the data includ-
ed in the Land Type Survey are discussed
in Fanourakis (2022).
From the above it is evident that a
significant amount of pedological clas-
sification and mapping has been conducted
worldwide. This data is used extensively
and successfully primarily for the optimisa-
tion of the utilisation of agricultural land
for crop production – so much so that
pedology is often incorrectly regarded as
part of agricultural science. It would be
obviously beneficial if this data could bene­
fit the engineering profession as well.
Objectives of this paper
This paper proposes a model to establish
the engineering soil classifications, in
particular the USCS and AASHTO, from
pedological data.
A research area was identified and the
soils in this area were classified according
to the South African Binomial System.
Samples were taken from the study area
and tested for their grading, plasticity and
chemical characteristics.
The grading results were used to
establish a procedure for determining the
engineering sized grading from the pedo-
logical grading sizes. This procedure was
required as the methods, sieve sizes and
soil separate boundaries used for pedologi-
cal purposes differ from those adopted for
engineering in South Africa. In addition,
the pedological soil separates are expressed
as a percentage (by mass) of the fraction of
the soil finer than 2 mm and not the entire
soil, as done for engineering purposes.
The grading, plasticity and chemical
properties were used to establish rela-
tionships that enable the determination
of the plasticity characteristics for the
pedological classification types included in
the research.
Based on the findings of this research
project, a mathematical model was
developed to determine the engineering
properties of a soil from the quantitative
and qualitative pedological data. The inputs
to this model are the physical and chemi-
cal properties of the soils of the various
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
horizons as established for pedological pur-
poses and supplied, for example, as part of
the modal profile data recorded in the Land
Type memoirs or stored in the databank
established by the Institute for Soil, Climate
and Water (LTSS 2021). The engineering
soil properties (output by the model) include
grading and plasticity properties, and the
USCS and AASHTO classification groups,
which may be used in the initial planning of
a wide range of soil engineering-related proj-
ects such as transportation route alignment
and township development. The model is
intended to supplement information at the
reconnaissance phase of proposed projects
by assisting in identifying relatively favour-
able or unfavourable soils.
The model was verified using data from
the study area, as well as from an area
located approximately 190 km from the
study area.
RESEARCH PROCEDURE
Research approach
A study area was selected as the basis for
collecting data, which, in turn, was appro-
priately analysed and was incorporated into
the proposed model to determine engineer-
ing properties from pedological data.
Research area
The area studied, which is located approxi-
mately 150 km northwest of the city of
Rustenburg in South Africa and is approxi-
mately 4 200 km2 in extent, lies to the
south of the confluence of the Crocodile
and Marico Rivers. The area was selected
for its diversity of soil types.
Geology
According to the Council for Geoscience
(1974) Geological Series Map 2426
Thabazimbi, Scale 1:250 000, the study
area may be generally divided into three
regions of varying geology.
The western region is underlain
predominantly by rocks of the Transvaal
Supergroup, together with rocks of the
Ventersdorp Supergroup and Bushveld
Igneous Complex.
The central region is underlain by
rocks of the Archaean Complex compris-
ing a granite basement with inliers of the
Swaziland Supergroup.
The northern region is underlain by
rocks of the Olifantshoek Supergroup
(Waterberg Group) with Post-Waterberg
diabase intrusions.
Tertiary and quaternary deposits are
present throughout the study area, occur-
ring mainly in the central region.
The soils investigated in this project
A number of soil types occur in the area,
differing from one another as a result
of different parent materials and/or the
different soil-forming processes to which
they had been subjected. The research was
confined to the dominant soil series which
adequately cover the range of soil types
occurring in the area. Although most of
the soils were transported, there were some
of residual origin.
Five soil forms were selected for the
research. These are the Hutton, Shortlands,
Valsrivier, Swartland and Arcadia forms,
which are respectively shown in the
typical photographs (with the associ-
ated horizon types) in Figures 1 to 5 in
Fanourakis (2012).
The correlations of the soil series select-
ed with the World Reference Base (WRB)
are included in Fanourakis (2012).
Fieldwork
The soils at 63 randomly selected sites
within the study area were classified
according to the South African Binomial
System (after MacVicar et al 1977), on the
basis of observing the soil profile in the
trial pits and the various horizon-based
series differentiation criteria. Such criteria
are detailed in Fanourakis (2012).
A total of 99 disturbed soil samples,
taken from the horizons of these soil pro-
files, were tested to determine the physical
and chemical properties that are used as
criteria for pedological classification. As
the physical and chemical properties of C
horizons are not reflected in the pedologi-
cal classifications of profiles which contain
C horizons, C horizons were not sampled.
Details of the methods of analysis and
results, as well as brief discussions on the
significance of each property determined
are given in the work of Fanourakis (1999).
In addition, the Atterberg Limits and
linear shrinkage of each sample were deter-
mined in accordance with TMH1 (1986).
The series encountered in the study
area, the percentage of the possible series
considered, the number of soil profiles
classified and the number of samples tested
for the determination of their plasticity
properties for the soils of each form are
shown in Table 1. All the possible series of
the five soil forms considered did not occur
in the study area.
3
Table 1 Percentage of series considered and number of profiles classified and tests conducted
Details of this research
Soil form
Percentage of
series considered
Hutton 25
Shortlands 22
Valsrivier 42
Swartland 17
Arcadia 8 17
ANALYSIS
Determining engineering
grading characteristics from
pedological grading
Engineering grading characteristics
The grading characteristics used as criteria
for the classification of a specific soil horizon
for engineering purposes according to the
USCS (USAEWES 1960) and the AASHTO
systems (HRB 1945) include the following:
QQ Masses of the fractions finer than
2.0 mm, 0.425 mm and 0.075 mm
(by mass) (P2.0 mm, P0.425 mm and
P0.075 mm), respectively.
QQ Alan Hazen’s Coefficient of Uniformity
(Cu) (Taylor 1948) and the Coefficient of
Curvature (Cc).
Grading data incorporated in
the Land Type memoirs
Grading data, determined by the physical
analysis of a soil horizon for series identifi-
cation purposes and presented in the format
used in the Land Type memoirs cannot be
used directly to classify a soil in accordance
with either of the engineering classification
systems above. This is due to the following:
QQ The method of particle-size analysis
used by the Institute for Soil, Climate
and Water to establish the grading
data for inclusion in the Land Type
memoirs is that described by Day
(1965), with modifications discussed in
Fanourakis (1999). This method differs
from methods employed by laboratories
performing grading analyses on soil
samples for engineering classification
purposes. These differences include the
adoption of sedimentation principles
(instead of sieve analysis) to determine
the grading of the fraction finer than
0.02 mm and 0.075 mm for pedological
and engineering purposes, respectively.
Furthermore, the pedological method
prescribes procedures for the removal of
organic matter (by oxidation with 30%
4 Volume 64 Number 1 March 2022
Number of profiles Number of plasticity
classified tests conducted
28 48
3 4
11 23
4 7
17
H2O2) and destroying of carbonates
(with 2 mol/l HCl).
QQ The pedological soil separate size
limits do not correspond to those used
to determine the engineering grading
characteristics.
QQ The masses of the pedological soil
separates are expressed as a percentage
of the mass of the fraction finer than
2.0 mm and not as a percentage of the
mass of the total sample as is done for
engineering purposes.
Method used for the determination
of pedological grading data
The physical analyses of the soil samples
included in this project to obtain those
grading characteristics necessary for series
identification in accordance with the South
African Binomial Classification System
(MacVicar et al 1977) were performed as
described by Day (1965), with modifica-
tions as discussed in Fanourakis (1999).
To determine the particle-size distribu-
tion of the fraction finer than 0.053 mm,
the hydrometer method was used in
this project whilst the pipette method is
employed by the Institute for Soil, Climate
and Water (LTSS 2021). Since Day (1965)
specifies both methods as alternatives, it
seems reasonable to assume that the results
obtained for this project would essentially
Table 2 Paired comparisons and corresponding t-test results
P2.0 mm
Sample number
Average
of tests
BC3A 99.3
AA1A 98.5
DB2B 99.1
AC3 98.0
AC2 98.9
Level of significance (P) 16.6 %
Difference significant?
be the same as would have been obtained
by the Institute for Soil, Climate and Water
on the same samples (LTSS 2021).
Procedure developed to estimate
engineering grading characteristics
from pedological data
A computer program to calculate values for
the grading characteristics which are used
for the classification of a soil for engineering
purposes from pedological grading data as
recorded in the Land Type memoirs was
written in BASIC (Beginner’s All-purpose
Symbolic Instruction Code) computer pro-
gramming language. A listing of the program,
as well as the linear interpolation-based pro-
cedure followed, is given in Fanourakis (1999).
The methods used by commercial labo-
ratories for determining the engineering
grading characteristics of a soil differ from
the method used by laboratories to deter-
mine the grading characteristics of a soil
for pedological purposes (Fanourakis 1999).
Hence, nine identical specimens of five
soil types, representative of the range of
soil types sampled, were submitted to nine
commercial engineering soils laboratories
for the determination of their engineering
grading characteristics. The grading analy-
ses determined by the different laboratories
and details on the test methods and dis-
persing agents used by each of the laborato-
ries are included in Fanourakis (1999).
Table 2 lists in parallel, for each soil
type, the mean value of each grading
characteristic (P2.0 mm, P0.425 mm,
P0.075 mm) determined by the nine com-
mercial engineering laboratories, on the
one hand, and the corresponding charac-
teristics estimated from pedological data
using the program developed, on the other.
The paired values were compared and
the null hypothesis assumed. The five per
cent significance level was adopted.
Grading characteristic
P0.425 mm P0.075 mm
Average Average
Estimated Estimated Estimated
of tests of tests
98.0 70.6 79.3 20.4 33.5
99.0 72.1 66.0 37.8 46.0
99.0 82.0 82.4 52.0 57.7
97.0 80.7 62.5 53.7 46.9
98.0 86.8 89.8 68.4 76.4
62.4 % 16.6 %
No No No
Journal of the South African Institution of Civil Engineering
 As a result of the fact that any significant
departures from the null hypothesis in either
of two directions (increase or decrease) were
important, the two-tailed t-test was applied
to the paired comparisons for each grad-
ing characteristic to establish the level of
significance. If the null hypothesis is true, the
critical t-value will follow a t-distribution with
(n-1) degrees of freedom. The significance
levels (P) which are the probabilities of differ-
ences of such magnitudes arising by chance,
as determined by each of the two-tailed
t-tests, are given in Table 2.
Referring to the levels of significance (P),
it is evident that the difference between the
paired values for each grading characteristic
is not significant. The value of the lowest
level of significance established (16.6%)
resulted from the t-tests performed on
paired values for the five soil types, in the
case of the percentages (by mass) finer than
2 mm and 0.075 mm. These values are, how-
ever, still 11.6% higher than the level of sig-
nificance (P) which would have justified the
rejection of the null hypothesis. Hence, it
was concluded for each of the three grading
characteristics that the difference between
the population means of each pair of values
was not significant and that the data was
consistent with the null hypothesis.
Nevertheless, the magnitude of the
differences between the paired values
would not have resulted in the allocation
of a different engineering soil classification
group symbol, regardless of whether the
laboratory or computer-estimated values
were used (provided that no plasticity was
exhibited by the soil, and the allocation of a
group symbol was dependent only on grad-
ing characteristics).
As the engineering grading character-
istics were estimated from the pedological
grading characteristics at a statistically
significant level, engineering grading
characteristics for the remaining samples
in the study area were not determined but
were instead estimated from the pedologi-
cal grading characteristics. It also appeared
reasonable to assume that the above proce-
dure is valid for determining engineering
grading characteristics of other soil types
and possibly any soil type from its pedo-
logical grading characteristics.
Regarding the Cu and Cc ‒ values of D10
and D30 could only be established from grad-
ing data, determined by the commercial labo-
ratories, in the case of only one (BC3A) of the
five soil types tested (in the case of the other
four soil types, D10 and D30 were less than
0.002 mm). Hence, on the basis of one sample,
Table 3 P
 ooled correlation coefficients and corresponding levels of significance for plasticity
characteristics
Level of significance (P%)
Soil form Pooled r
Minimum Maximum Mean
Hutton (Ferralsols and Arenosols)1 0.863 4.7E-11 5.3E-04 1.8E-04
Shortlands, Valsrivier and Swartland
0.866 3.7E-12 4.7E-09 2.2E-09
(Luvisols)1
Arcadia (Vertisols)1 0.707 5.7E-02 5E-01 2.2E-01
1 General equivalent according to the WRB Classification System (IUSS-WRB 1998; 2006; 2007; 2015).
the degree of confidence with which Cu and
Cc can be estimated from the pedological
Pedological Data
data could not be determined statistically.
However, practically, the Cu and Cc values
may be required for classification according Pedological grading
to the USCS, in the case of only two (GW and and chemical
characteristics
SW) of the 15 possible groups comprising the
system. These are not required for classifica-
tion according to the AASHTO system. Pedological Pedological
chemical grading
characteristics characteristics
Determining plasticity
characteristics from the soil series
Fanourakis (2012) established relationships Estimated
engineering
for determining the engineering plasticity grading
characteristics ‒ namely liquid limit (LL), characteristics
plasticity index (PI) and linear shrinkage
(LS) ‒ of soils deriving from a range of
Estimated plasticity
pedological classes of the South African characteristics
Binomial System (after MacVicar et al 1977)
from their physical and chemical properties
which are determined for pedological clas- Engineering
classification
sification purposes. More specifically, in the
case of each soil form considered, each of
the plasticity properties (LL, PI and LS) was USCS AASHTO
correlated with the amount of magnesium classification classification
group group
present in the clay-size portion of the frac-
tion of the sample finer than 0.425 mm.
For each soil group, Fisher’s Figure 1 F low diagram illustrating the
Z-transformation technique was used to structure of the model
combine independent correlation coef-
ficients, which are significant at the five per The pooled correlation coefficients and
cent probability level, in order to obtain an corresponding levels of significance are
estimate of the average (pooled) correla- summarised in Table 3. The grouping of
tion coefficient. This procedure took into the Shortlands, Valsrivier and Swartland
account each correlation coefficient and soil forms is justified in Fanourakis (2012).
the corresponding number of sets of values
on which it was based.
Significance tests were carried out for THE MATHEMATICAL MODEL
each relationship established using the
Student’s t-distribution (Gosset 1908) to Components of the model
determine the probability that the correla-
tion coefficient could have arisen by chance Structure of the model
in a sample of the size dealt with. All the A flow diagram illustrating the general
relationships established for estimating the structure of the model is shown in Figure 1.
plasticity characteristics of the soils were
highly significant (P < 1%). These relation- Pedological input data
ships are represented and discussed in The pedological data, which is the input
detail in Fanourakis (2012). data, consists of the following:

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 5
QQ The soil form Table 4 Actual and estimated classification group symbols for soils of the Hutton form
QQ The quantitative grading characteristics
USCS USCS AASHTO AASHTO
of the soil sample as given for pedologi- Designation Soil series
actual estimated actual estimated
cal purposes AC1A Hu Shorrocks SC SC A-6 A-6
QQ The exchangeable magnesium (in me %),
AC1B Hu Shorrocks SC SC A-6 A-6
included in the quantitative chemical
characteristics of the soil as given for AC4A Hu Makatini SC SC A-4 A-6

pedological purposes. AC4B Hu Makatini CL CL A-6 A-6

BB1A Hu Portsmouth SM SM A-2-4 A-2-4
Engineering soil properties output BB1B Hu Portsmouth SM SM A-2-4 A-2-4
BC1A Hu Shigalo SM SM A-2-4 A-2-4
Engineering grading characteristics
BC1B Hu Shigalo SM SM A-4 A-4
The engineering grading characteristics
of the soil sample are estimated from the BC2A Hu Portsmouth SM SM A-2-4 A-2-4
pedological grading characteristics. BC2B Hu Portsmouth SM SM A-4 A-4
BC3A Hu Portsmouth SM SM A-2-4 A-2-4
Plasticity characteristics BC3B Hu Portsmouth SM SM A-4 A-4
In the case of each soil form, the LL, PI and
BD1A Hu Shigalo SM SM A-4 A-4
LS are estimated using the highly signifi-
CB2A Hu Shorrocks SM-SC SC A-4 A-4
cant mathematical relationships obtained
between these properties and the physical CB2B Hu Shorrocks SC SC A-6 A-6
and chemical soil properties such as the per- CB3A Hu Vergenoeg SM SM A-4 A-4
centage finer than 0.425 mm (P0.425 mm), CB3B Hu Vergenoeg SM SM A-4 A-4
the clay content (P0.002 mm) and the quan- CB4A Hu Shorrocks SM SM A-4 A-4
tity of the exchangeable magnesium cations
CB4B Hu Shorrocks SC SC A-6 A-6
(in me %) (Fanourakis 2012).
Fanourakis (2012) proposed that soils CB5A Hu Portsmouth SM SM A-4 A-4

of the Hutton form with a clay content CB6B Hu Shorrocks CL CL A-6 A-6
less than 12% are automatically designated CB7A Hu Mangano SM SM A-4 A-4
as non-plastic. This 12% was the average CB7B Hu Mangano SM SM A-2-4 A-2-4
clay content of 18 Hutton soils that were
CD1A Hu Doveton SM-SC SC A-4 A-4
non-plastic and had clay contents ranging
CD1B Hu Doveton CL CL A-6 A-6
from 6% to 18%. In this paper it is proposed
that the relationships pertaining to soils of DA10A Hu Makatini CL CL A-4 A-6
the Hutton soil form should not be applied DA11B Hu Shorrocks CL CL A-6 A-6
to soils with a clay content equal to or less DA12A Hu Shigalo CL CL A-4 A-6
than 18%, as these were not included in DA12B Hu Shigalo CL CL A-6 A-6
the establishment of the relationships. The
DA14B Hu Shorrocks CL CL A-6 A-6
estimation of plasticity characteristics from
DA17A Hu Msinga CL CL A-6 A-6
pedological data is discussed in detail in
Fanourakis (2012). DA17B Hu Msinga CL CL A-6 A-6
These relationships to determine DA19A Hu Msinga SM SM A-4 A-4
plasticity characteristics were applied using DA19B Hu Msinga SC SC A-6 A-6
data which fell within the range on which DA20B Hu Msinga SC SC A-4 A-4
the relevant relationships were established.
DA6A Hu Shorrocks SM SM A-4 A-4

Engineering classification of soils DB2A Hu Makatini SC SC A-4 A-4

The soil is classified according to the two DB2B Hu Makatini CL CL A-6 A-6
most commonly employed engineering soil DB6A Hu Doveton SC SC A-4 A-6
classification systems, namely the USCS DB6B Hu Doveton CL CL A-6 A-6
(USAEWES 1960) and the AASHTO sys-
DB8A Hu Shorrocks CL CL A-6 A-6
tem (HRB 1945).
DB8B Hu Shorrocks SC SC A-6 A-6

Limitations of the model DC1A Hu Makatini CL CL A-6 A-6

This research entailed the investigation DC1B Hu Makatini CL CL A-6 A-7-6
of soils occurring in a particular area. DC2A Hu Msinga SM SM A-4 A-4
Not all the series encompassed by the DC3A Hu Shorrocks SC SC A-4 A-4
forms investigated occur in the study area.
DC3B Hu Shorrocks CL CL A-6 A-6
Furthermore, the behaviour of soils of the
Note: Blue indicates two class differences.
series researched which have a chemical and

6 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Table 5 A
 ctual and estimated classification group symbols for soils of the Shortlands, Valsrivier mineralogical composition differing from
and Swartland forms that of the soils included in the research is
USCS USCS AASHTO AASHTO not known. Hence, the plasticity relation-
Designation Soil series ships could not be applied in such cases.
actual estimated actual estimated
AA1A Sd Kinross SC SC A-6 A-6
CC2A Sd Ferry CL CL A-4 A-6
CC2B Sd Ferry CL CL A-6 A-6 VERIFICATION OF THE MODEL
DA3B Sd Kinross SC SC A-6 A-6
CC1A Sw Reveillie CL CL A-4 A-6 Verification sources
CC1B Sw Reveillie CL CL A-6 A-6
The model was verified by being applied
CC5A Sw Skilderkrans CL CL A-7-6 A-7-6
using pedological data to determine the
CC5B Sw Skilderkrans CL CL A-6 A-7-6
engineering properties of soils in the
study area.
CC6A Sw Skilderkrans CL CL A-7-6 A-6
In addition, the model was verified
CC6B Sw Skilderkrans CL CL A-6 A-6
using data from a pedological and geo-
DB5A Sw Skilderkrans CL CL A-6 A-6
technical report pertaining to an area
AA2A Va Waterval CL CL A-6 A-6
in Lephalale (formerly Ellisras), located
AA2B Va Waterval CH CL A-7-6 A-7-6
approximately 190 km from the study area.
BB2A Va Sunnyside CL CL A-6 A-6
BB2B Va Sunnyside CL CL A-7-6 A-7-6
Verification using data
CC3A Va Marienthal CL CL A-4 A-6
from the study area
CC3B Va Marienthal MH CH A-7-5 A-7-6
CC4B Va Valsrivier MH MH A-7-5 A-7-5
Data used
DA18A Va Waterval SC SC A-6 A-6
The USCS and AASHTO classifications for
DA18B Va Waterval CL CL A-6 A-6
each of the samples tested as part of this
DA7A Va Waterval CL CL A-4 A-6
research project were determined by using:
DA7B Va Waterval CL CL A-7-6 A-7-6
QQ the estimated engineering grading
DA7B2 Va Waterval CL CL A-7-6 A-7-6
characteristics and the plasticity results
DA8A Va Waterval SC SC A-4 A-6
obtained from standard engineering
DA8B Va Waterval CL CL A-7-6 A-7-6
laboratory tests, on those samples, and
DA8B2 Va Waterval CL CL A-7-6 A-7-6
QQ the engineering grading and plasticity
DB3A Va Marienthal CL CL A-7-6 A-7-6
characteristics by applying the model to
DB3B Va Marienthal CH CH A-7-6 A-7-5
the quantitative physical and chemical
DB4A Va Lindley ML CL A-7-6 A-7-6
pedological data of the same samples.
DB4B Va Lindley CL CL A-7-6 A-7-6
One soil of the Hutton form and three of
DC4A Va Lindley CL CL A-6 A-6
the Arcadia form were excluded, as they
DC4B Va Lindley CL CL A-7-6 A-7-6
had physico-chemical properties outside the
DD3A Va Lindley CL CL A-6 A-6
ranges pertaining to the relevant plastic rela-
DD3B Va Lindley CH CH A-7-6 A-7-6
tionships. The engineering soil classification
Note: Green indicates one class difference. Blue indicates two class differences. group symbols mentioned above are shown in
parallel, for the 95 soils of the different forms
Table 6 Actual and estimated classification group symbols for soils of the Arcadia form included in the study area, in Tables 4 to 6.
USCS USCS AASHTO AASHTO
Designation Soil series Results and discussion
actual estimated actual estimated
AC3 Ar Arcadia SC SC A-7-5 A-7-5 Table 7 shows the a summary of soil form
CB1 Ar Arcadia CH CH A-7-6 A-7-6 specific results of the accuracy of the
CD2 Ar Arcadia CH CH A-7-6 A-7-6
estimated classification group symbols,
according to the AASHTO and USCS sys-
DA13 Ar Arcadia CH MH A-7-6 A-7-5
tems, for the soils considered.
DA15 Ar Arcadia SC SC A-7-5 A-7-5
Considering the soils of all the series
DA2 Ar Arcadia CH CH A-7-6 A-7-6
included in this research, a total of six of
DA4 Ar Arcadia SC SC A-7-5 A-7-5
the 95 (6%) estimated USCS group symbols
DA5 Ar Arcadia SC SM A-2-7 A-2-7
differed from actual symbols of the same
DA9 Ar Arcadia CH CH A-7-6 A-7-6
soils by one group. No samples differed by
DB1 Ar Arcadia CH CH A-7-5 A-7-5
more than one group. In cases where either
DB7 Ar Arcadia CH CH A-7-5 A-7-5
the actual or estimated symbol was included
DB9 Ar Arcadia MH CH A-7-5 A-7-5
in the other, this was considered not to
DD1 Ar Arcadia CH CH A-7-6 A-7-5
be different. For example, the actual and
DD2 Ar Arcadia CH CH A-7-5 A-7-5
estimated USCS symbols assigned to sample
Note: Green indicates one class difference. CB2A were SM-SC and SC, respectively.

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 7
Table 7 Summary of accuracy of classification group estimations according to soil forms
Soil form
Difference in group
classification classes Shortlands,
Classification system Hutton Valsrivier and Arcadia Total (in row)
between actual
and estimated (47 samples) Swartland (14 samples)
(34 samples)

None (correct) 47 31 11 89
USCS
1 Group 0 3 3 6

None (correct) 42 25 12 79

AASHTO 1 Group 0 2 2 4

2 Groups 5 7 0 12

In the case of the AASHTO classifica-
tion, the estimated and actual group sym-
bols of a total of 16 samples (17%) differed.
Four of these differed by one class and 12
differed by two classes. In this analysis, the
classes of A-7-5 and A-7-6 were considered
to be different classes (although they
comprise the A-7 class). The four soils
which differed by one class were such soils.
None of the soils differed by more than two
classes. Furthermore, nine of the 12 soils
that differed by two classes had actual and
estimated groups of A-4 and A-6, defined
by a PI of 10 max or 11 min, respectively.
A total of four soils differed in subclass
within the A-7 class (A-7-5/A-7-6).
Verification using soil data from
areas outside the study area
Area considered
The validity of the model was also tested
using soil data of the Lephalale area in
the Limpopo Province which, according
to the map of the pedology of southern
Africa, after Harmse and Hattingh (1985),
generally comprises similar soils to those
researched but is located approximately
190 km northwest of the study area.
Pedological and engineering data of
the soils occurring along an approxi-
mately 38 km section of the Steenbokpan‒
Stockpoort Road (Road 175 which is
situated in this area) were obtained from
the Pedological Research Institute and the
(former) Roads Department of the Transvaal
Provincial Administration (TPA 1987),
respectively. The location of this road in
relation to Lephalale is shown in Figure 2.

Figure 2 The location of the Steenbokpan–Stockpoort Road in relation to Lephalale

8 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Table 8 Estimated and actual AASHTO classification groups at various chainages along the Steenbokpan–Stockpoort Road
Chainage Pedological
(km) profile
1.1 SP 26
2.6 SP 4
4.2 HK 6
11.2 HK 19
12.7 HK 5
16.4 HK 20
Data obtained
The pedological data acquired comprised
six soil profiles which were recorded along-
side or very close to this section of the road
as part of a pedological mapping exercise
(conducted to a 1:50 000 scale), as well
as the pedological physical and chemical
data of the soils constituting these profiles
(IPR 1989).
The engineering data comprised the
pavement and gravel pit design sheets
which were compiled using data gath-
ered as part of a detailed investigation
conducted along this section of the road.
The pavement design sheets, which were
drafted using a standard form, included
data on the colour, depth, grading, plastic-
ity, bearing strength (CBR) and AASHTO
classification of the natural soils occurring
on the road centreline at 200 m intervals.
The gravel pit design sheets included data
such as the material description, profile
depth, grading characteristics, plasticity
characteristics, AASHTO classification
and bearing strength characteristics of the
material occurring in nine trial pits exca-
vated in each proposed borrow area (TPA
1987). The USCS classification groups were
not given.
The area adjacent to this road is char-
acterised by red soils of the Hutton and
Bainsvlei soil forms and yellow-brown soils
of the Clovelly and Avalon forms. Soils
of the Valsrivier, Swartland and Arcadia
forms which were included in the research
are generally located more than 2 km from
the road.
Application of the model
The model was used to determine the
AASHTO classification of the soils
represented by the six profiles along the
road route from the pedological data of
these soils.
Of the soils comprising the six profiles,
only the properties of the Hutton soils had
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
Depth (mm)
Soil form and
series
Pedological Engineering
Clovelly Gutu 0–500 0–650
Hutton Portsmouth 0–1 300 0–700
Hutton Moriah 0–1 850 0–800
Bainsvlei Maanhaar 0–670 0–760
Hutton Bontberg 0–1 100 0–670
Avalon Windmeul 0–290 0–600
been researched for the purposes of estab-
lishing the model. However, the Bainsvlei
soil profile consists of a Hutton soil profile
underlain by a soft plinthic (ferruginised)
horizon. Furthermore, the Clovelly profile,
like the Hutton profile, has an intact
structure and clays present are mainly
of the 1:1 kaolinitic type, but is yellow in
colour and not red as in the case with the
Hutton soil profile. The Avalon soil profile
consists of a Clovelly soil profile which
is underlain by a soft plinthic horizon. In
other words, the Hutton, Bainsvlei, Clovelly
and Avalon soil forms are all characterised
by Apedal B horizons.
Therefore, although the properties
of the soils of the Bainsvlei, Avalon and
Clovelly forms were not investigated as part
of the project on which the model is based,
the section of the model applicable to the
soils of the Hutton form was applied using
pedological data of the Clovelly form soils
and soils overlying the soft plinthic horizon
of the Bainsvlei and Avalon forms to estab-
lish whether the engineering properties of
these soils would be the same as those of a
Hutton soil with an identical physical and
chemical composition.
Discussion and results
Subgrade properties
The AASHTO classification groups
determined for the horizons in each
profile (estimated) using the model, on the
one hand, and determined as part of the
detailed centreline engineering survey at
the appropriate chainages on the other, are
given in Table 8.
With reference to Table 8, the profile
depths according to the pedological and
engineering profiles differ.
Nevertheless, as indicated in Table 8,
the estimated and determined AASHTO
classification group symbols are essentially
the same at each chainage. Where, for
AASHTO classification group
Estimated Actual
A-2-4 A-2-4
A-2-4/A-2-6 A-2-4
A-2-4 A-2-4/A-3
A-2-4 A-2-4/A-3
A-2-4 A-2-4
A-2-4 A-2-4
example in the case of HK 6 and HK 19,
two groups were given as the actual, these
included the estimated group (A-2-4).
Similarly, in the case of SP 4, where two
groups were given as the estimated, these
included the actual group.
Gravel pits
Thirteen potential borrow pits located
along the road route were investigated as
part of the detailed engineering survey.
Nine test pits were excavated in each of
the 13 potential borrow areas. The profiles
in all of these borrow areas comprised
calcrete or red and/or brown soil which
is underlain by calcrete or ferricrete (TPA
1987). In view of the fact that the pedologi-
cal test pits profiled were located such that
the range of soils occurring in the area
was represented by these profiles, none of
the pedological test pits were specifically
placed in any of the borrow pits. However,
four of the six pedological test pits profiled
alongside the road were situated within
a distance of 0.6 km and 2.0 km from
borrow pits.
The AASHTO classification groups
in Table 9 were determined by applying
the model to pedological data obtained
from the four pedological profiles in the
vicinity of the potential borrow areas, and
the corresponding values were determined
on samples from the test pits in the four
potential borrow areas by conventional
engineering test procedures.
A comparison of the data in Table 9
revealed that the actual and estimated
AASHTO classification groups, in the two
parallel lists, were essentially the same.
Where, for example in the case of HK 19
and HK 5, two groups were given as the
actual, these included the estimated group.
Similarly, in the case of SP 4, where two
groups were given as the estimated, these
included the actual group. In any case,
the minor differences could probably be
9
Table 9 Estimated and actual AASHTO classification groups of material in four test pits occurring alongside the Steenbokpan–Stockpoort Road
Nearest Depth (mm) AASHTO classification group
Chainage Number of
pedological
(km) samples tested Pedological Engineering Estimated Actual
profile

3.24 3 SP 4 0–1 300 0–1 200 A-2-4/A-2-6 A-2-4

9.0 9 HK 19 0–670 0–1 500 A-2-4 A-2-4/A-3

14.0 7 HK 5 0–1 100 0–1 500 A-2-4 A-2-4/A-2-6

17.25 9 HK 20 0–290 0–1 300 A-2-4 A-2-4

attributed to the fact that the plasticity
properties of the soils in the four pedologi-
cal profiles might have differed somewhat
from those of the soils in the test pits since
the pedological profiles and test pits were
not identically located.
It is also important to note that,
although the model was developed from
test results of, amongst others, red soils of
the Hutton form but not from test results
of red soils of the Bainsvlei form or yellow-
brown soils of the Clovelly and Avalon
forms (which were not included in the
research), the mathematical model appears
to be valid for these soils as well. This is
probably because all the above-mentioned
soil forms have Apedal B horizons.
However, it must be noted that, in the
model, soils of the Hutton form with a clay
content less than 18% are designated as non-
plastic. Therefore, since the soils in the area
traversed by this road generally have a low
clay content, the validity of the plasticity sec-
tion of the model could not be confirmed.
CONCLUSIONS
The model developed
and its verification
A mathematical model was successfully
developed that enabled the determination
of engineering grading and plasticity char-
acteristics, as well as classification (USCS
and AASHTO) from pedological data.
This model was verified using data from
the study area, as well as data from an area
located approximately 190 km from the
study area.
Verification using data
from the study area
The engineering grading characteristics
of soils from the study area were deter-
mined to a statistically significant degree
of accuracy (using the five per cent level
of significance).
The plasticity characteristics of soils
from the study area were statistically
significantly estimated from the physico-
chemical data of soils, for a range of
pedological groups. All the relationships
established for estimating the plasticity
characteristics of the soils of the Hutton
form (Ferralsols and Arenosols according
to the WRB Soil Classification System),
Arcadia form (Vertisols according to the
WRB system) and Shortlands, Valsrivier
and Swartland forms (Luvisols according
to the WRB system) were highly significant
(P < 1%) (Fanourakis 2012).
Regarding the estimated USCS and
AASHTO classification group symbols of
soils from the study area, of the 95 soils
included in the project, in total, 12% of the
estimated classification group symbols dif-
fered from the actual classification group
symbols. Furthermore, only 6% of the clas-
sification groups differed by two groups.
However, only in 2% of the soils (CC3B and
DA13) did the estimated and actual clas-
sification group symbols, according to both
the above engineering soil classification
systems, differ. However, if the estimated
PIs of CC3B and DA13 were 1% lower and
1% higher, respectively, the estimated and
actual USCS symbols would have been the
same. From all the above results it can be
concluded that the estimated engineering
classification group symbols, determined
by applying the model, provide a suf-
ficiently accurate indication of the probable
engineering behaviour of these soils.
Verification using data from
outside the study area
The estimated AASHTO classification
groups of soils of the Hutton, Bainsvlei,
Clovelly and Avalon forms, which occur
adjacent to a road located approximately
190 km north of the study area, were
determined using physical and chemical
pedological data and the section of the
model pertaining to the soils of the Hutton
form. These estimated classification groups
did not differ significantly from the actual
classification groups which were deter-
mined from soils that had been sampled
from the same locations or nearby as part
of a detailed engineering survey.
Application of the model
and further research
The findings of this research are valid
for soils of the pedological classifications
investigated and their particular physical,
chemical and mineralogical proper-
ties only, since the research project was
intended to serve as a pilot study. The
indications are that additional similar
studies covering soils of the pedological
classifications researched but with different
chemical properties, as well as soils with
other pedological classifications with vary-
ing chemical properties, could undoubtedly
lead to the extension of this work and
improvement of its universality.
This model is intended to assist in the
supplementing of information in the recon-
naissance stages of certain projects.
Finally, it is hoped that this research
has succeeded in emphasising the inter-
relationships between pedogenesis and
the engineering behaviour of soils, and in
suggesting an approach for the interpreta-
tion of pedological data for engineering
purposes.
References
Calitz, F 2021. Proposed system to facilitate use
of pedological information in preliminary
stage geotechnical investigations. PhD Thesis.
Potchefstroom: North-West University.
Council for Geoscience 1974. Geological Series Map
2426 Thabazimbi. Scale 1:250 000. Pretoria:
Government Printer.
Day, P R 1965. Physical and mineralogical properties,
including statistics of measurement and
sampling. In Black, C A, Evans, D D, White, J L,
Ensminger, L E & Clark, F E (Eds.), Methods of Soil
Analysis – Part 1. Madison, WI: American Society
of Agronomy.
FAO (Food and Agriculture Organization of the United
Nations) 2021. Soils Portal, Data Hub. www.fao.
org/soils-portal/soil-survey/soil-classification/
world-reference-base/en.

10 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Fanourakis, G C 1999. Engineering soil properties from
pedological data. DTech (Eng) Thesis. University of
Johannesburg.
Fanourakis, G C 2012. Estimating soil plasticity properties
from pedological data. Journal of the South African
Institution of Civil Engineering, 54(2): 117‒125.
Fanourakis, G C 2022. Utilising land type data for
geotechnical investigations. Journal of the South
African Institution of Civil Engineering, 64(1): 12‒24.
Gosset, W S 1908. The probable error of a mean.
Biometrika, 6(1): 1‒25.
Harmse, H J Von M & Hattingh, J M 1985. The
pedological profile classes. In Brink, A B A (Ed.),
Engineering Geology of Southern Africa, Vol 4, 272–273.
HRB (Highway Research Board) 1945. The United
States Bureau of Public Roads Classification System.
Proceedings, 25th Annual Meeting of the HRB,
Washington, DC.
IPR (Institute for Pedological Research) 1989.
Pedological maps, profile descriptions and physical
and chemical properties of the soils in the area
covered by the Steenbokpan, Hardekraaljie and
Stockpoort 1:50 000 Quarter Degree Sheets.
Potchefstroom University for CHE.
IUSS-WRB (International Union of Soil Science
Working Group (IUSS)–World Reference Base
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
for Soil Resources (WRB)) 1998. A framework
for international classification, correlation and
communication. World Soil Resources Report
No 84. Rome: Food and Agriculture Organization of
the United Nations (FAO).
IUSS-WRB 2006. A framework for international
classification, correlation and communication.
World Soil Resources Report No 103. Rome: FAO.
IUSS-WRB 2007. World Reference Base for Soil
Resources 2006. First update 2007. A framework
for international classification, correlation and
communication. World Soil Resources Report
No 103. Rome: FAO.
IUSS-WRB 2015. World Reference Base for Soil
Resources 2014. Update 2015. International soil
classification system for naming soils and creating
legends for soil maps. World Soil Resources Report
No 106. Rome: FAO.
LTSS (Land Type Survey Staff) 2021. Land types of
South Africa: Digital map (1:250 000 scale) and soil
inventory databases. Pretoria: Institute for Soil,
Climate and Water, Agricultural Research Council.
https://www.arc.agric.za/arc-iscw/Pages/ARC-
ISCW-Homepage.aspx.
MacVicar, C N, De Villiers, J M, Loxton, R F, Verster,
E, Lambrechts, J J N, Merryweather et al 1977. Soil
classification: A binomial system for South Africa.
Scientific Bulletin No 390. Pretoria: Department of
Agriculture Technical Services.
SCWG (Soil Classification Working Group) 2018. Soil
classification: A natural and anthropogenic system
for South Africa. Pretoria: Institute for Soil, Climate
and Water, Agricultural Research Council.
Taylor, D W 1948. Fundamentals of Soil Mechanics
(13th printing, December 1963) p 9.
TMH 1 1986. (Technical Methods for Highways).
TMH 1 1986: Standard Methods of Testing
Road Construction Materials, 2nd ed. Pretoria:
Department of Transport.
TPA (Transvaal Provincial Administration) 1987.
Centre line survey for the proposed Stockpoort‒
Steenbokpan Road (Road 175). Conducted by Van
Wyk and Louw Inc for the TPA Roads Department,
Job No 2420, Ellisras District, Drawing Nos PRS
87/171/1Sp, 2Sp, 106–109, 118, 120–122.
USAEWES (United States Army Engineer Waterways
Experiment Station) 1960. The Unified Soil
Classification System. Technical Memorandum
No 3-357. Appendix A, Characteristics of soil groups
pertaining to embankments and foundations, 1953;
Appendix B, Characteristics of soil groups pertaining
to roads and airfields. Vicksburg, MS: USAEWES.
11
TECHNICAL PAPER Utilising Land Type data for
Journal of the South African
Institution of Civil Engineering
geotechnical investigations
ISSN 1021-2019
Vol 64 No 1, March 2022, Pages 12–24, Paper 1311 G C Fanourakis

PROF GEORGE FANOURAKIS (FICE, FSAICE) is

Associate Professor in the Department of
Civil Engineering Technology at the (now)
University of Johannesburg, which he joined
around 28 years ago after leaving the
employ of Jones and Wagener (Pty) Ltd
Consulting Engineers. He holds the degrees
of MSc(Eng) and DTech(Eng), and is a
Chartered Civil Engineer and Fellow of the Institution of Civil Engineers
(UK). He is also a Fellow of the South African Institution of Civil
Engineering (SAICE), an Honorary Fellow (and Past-President) of the
Institute of Professional Engineering Technologists (IPET) and Member of
the Soil Science Society of Southern Africa. His professional participation
further includes membership of three South African Bureau of Standards
(SABS) committees, membership of COM 9: Dissemination of Knowledge of
the fib (fédération internationale du béton), and membership of TC:306
(Geo-Engineering Education) of the International Society of Soil Mechanics
and Geotechnical Engineering (ISSMGE).
Contact details:
Department of Civil Engineering Technology
University of Johannesburg
PO Box 17011
Doornfontein
2028
South Africa
T: +27 11 559 6416
E: georgef@uj.ac.za
Keywords: land type survey, soil surveys, pedology, soil mapping,
geotechnical investigations
In South Africa, a Land Type Survey (LTSS) has been conducted of the entire country. The information
available from the LTSS is readily and inexpensively made available to the public (in the form of
maps and accompanying memoirs). This survey includes information on the climate, terrain and
soils. The soil information given includes qualitative data (such as pedological classifications, soil
profiles and soil types), as well as quantitative physical and chemical properties of representative
soils tested. Such information is used primarily for agricultural purposes. This paper proposes
an approach for inferring information, of relevance in reconnaissance geotechnical engineering
surveys, from Land Type data, to supplement geotechnical investigation data. The proposed
procedure was verified using actual information from a geotechnical report. This investigation
indicated that general engineering properties can be inferred from qualitative pedological data. In
addition, statistically significant engineering properties (including the grading, Atterberg Limits and
the USCS and AASHTO classifications) were determined from quantitative pedological data, in 86%
of cases, for the soils considered. It is evident from this research that data from the Land Type survey
data can be successfully used for geotechnical investigation purposes for certain developments
(such as roads and townships) where the depth of interest is relatively shallow.
INTRODUCTION information (in a standard format), that has
already been compiled for a number of coun-
Pedological classification tries. This includes the New Generation Soil
systems and databases Property Maps for Africa (ISRIC 2021).
There are various definitions of pedology. Therefore, a significant amount of
Essentially, pedology is the study of soils as pedological classification and mapping
naturally occurring phenomena, including has been conducted worldwide. Although
their formation, composition, classification this data is used primarily for agricultural
and distribution. purposes, such information, particularly
A number of pedological soil classifica- that on soils and terrain, could be used for
tion systems have been established and used geotechnical engineering investigations.
as a basis for mapping soils. Such systems,
which include an internationally accepted The Land Type Survey
system and many national systems, are listed The Land Type Survey, which was initi-
in Fanourakis (2012) and Fanourakis (2022). ated in 1972 and was completed in 2002
The commonly used South African (Patterson et al 2015), comprises a publicly
Pedological Soil Classification System available completed survey of the entire
is titled “Soil Classification: A Binomial South Africa, which includes information
System for South Africa” (MacVicar et al on the soils, terrain, climate and geology.
1977). An overview of this classification is This survey was conducted by the Institute
given in Fanourakis (2012). of Soil, Climate and Water (formerly the
Soil and Irrigation Research Institute) of
Pedological databases the Agricultural Research Council.
Various international, regional and The soil data included in the Land Type
national databases comprising pedological, Survey of South Africa (LTSS 2021) includes
climatic, topographical and other informa- the pedological properties and pedological
tion have been developed, as detailed in classification in accordance with the South
Fanourakis (2012). African Binomial Classification System
The South African database, which (after MacVicar et al 1977).
contains information for the entire country, The data acquired in the course of the
is the Land Type Soil Survey (LTSS 2021). Land Type Survey is published in the form
The Land Type Survey has been incor- of maps (at a scale of 1:250 000), known
porated into the SOTER (Soil and Terrain) as Land Type maps, and accompanying
world database, which includes maps and soil documents which are referred to as Land

Fanourakis GC. Utilising Land Type data for geotechnical investigations.

12 J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #1311, 13 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64n1a2
Type memoirs. At the time of publication
of this paper, the cost of a Land Type map
on paper (on a topo-cadastral background)
was R100.00 and on semi-transparent matt
film (without a topo-cadastral background)
was R350.00. The cost of memoirs, which
are currently mainly available on CD (as
opposed to a hardcopy book), was R75.00
each. In addition, an optional CD containing
an explanation of the contents of any par-
ticular map is available at a cost of R75.00.
Objectives of this paper
Despite the different depths of soil which are
of concern in pedology and in engineering,
information gathered by pedologists is also
of potential use to engineers. Translating the
classification system on which pedological
maps, such as the Land Type maps, are based
into engineering terms, even if only for the
uppermost 1.2 m of the soil profile, may be
of significant value in engineering projects
concerned with transportation links, canals,
township development and the location of
natural construction materials (Harmse &
Hattingh 1985).
Paige-Green and Turner (2007) generally
discussed how the information from the
Land Type Survey could potentially assist,
Figure 1 Example of a Land Type inventory – Ea26 (LTSS 2021)
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
specifically, in road engineering surveys.
Furthermore, Calitz (2021) developed a
detailed Soils Effects Grouping (SEG) system
to convert pedological information to infor-
mation useful in geotechnical assessments.
The approach used by Calitz (2021) is differ-
ent to that discussed in this paper.
The primary objective of this paper is to
demonstrate the extent to which Land Type
data can be used in geotechnical investiga-
tions, particularly at the reconnaissance
stages, which are undertaken at the planning
stages of proposed developments, where the
depth of interest is relatively shallow (e.g. for
roads and township developments). The ter-
rain and soil information inferred from the
LTSS would typically be used together with
other information, such as geological maps,
aerial photographs, topographical maps and
technical reports (if available) to effectively
plan detailed soil surveys. In this paper, this
was illustrated by utilising the LTSS data
pertaining to a particular road route to infer
information or estimate properties applicable
to the reconnaissance geotechnical investiga-
tion. The inferred information was then veri-
fied by being compared to the data from the
actual engineering geotechnical investigation
report that was compiled for the same road.
In achieving this objective, an overview
of the information that is incorporated in
the Land Type Survey is given.
THE LAND TYPE SOIL SURVEY
The identification of Land Types
Areas known as Land Types were identified
and delineated on a 1:250 000 scale, such
that each Land Type displayed a marked
degree of uniformity with regard to terrain
form, soil pattern and climate. The Land
Types were initially delineated on 1:50 000
quarter degree sheets, each comprising
a 1:250 000 sheet, and then the bounda-
ries were transferred onto the relevant
1:250 000 sheets (LTSS 2021).
The establishment of Land
Type inventories
Approximately 7 070 Land Types were
established, based on approximately
400 000 soil observations, which translate
to approximately one observation per
300 ha (Patterson et al 2015).
A Land Type inventory was established
for each delineated unit (Land Type). The
Land Type inventories are included in the
13
relevant memoirs. Each Land Type inven-
tory contains information on the terrain,
pedology, climate and geology of that Land
Type. An example of a Land Type inven-
tory is shown in Figure 1 (on page 13).
With reference to Figure 1, the follow-
ing information regarding soil and terrain
is included in the inventory:
QQ The Land Type designation (shown in
the blue rectangle), corresponding with
that on the map unit (e.g. Ea26).
QQ The area of the Land Type unit (shown
in the olive rectangle).
QQ A terrain type and terrain form sketch
(profile sketch) in the red rectangle.
This indicates the terrain type (e.g. A3)
which is based on estimates of the slope
and local relief, according to standard
ranges, representative of the terrain
characterising the Land Type. The sec-
tion (profile) which is representative of
the terrain of the Land Type comprises
(numbered) terrain units. A terrain
unit is any part of the land surface
with a homogeneous form and slope
(LTSS 2021).
QQ Classification of the soils occupying
the Land Type according to the South
African Binomial Classification System
(MacVicar et al 1977) (shown in the
green rectangle).
QQ Clay content ranges for each soil horizon
type (shown in the yellow rectangle).
QQ The textural class of each horizon
(shown in the light blue rectangle).
Figure 2 Example of a modal profile description (LTSS 2021)
14
QQ Depth range of soil/rock horizons of each
soil form (shown in the purple rectangle).
QQ The depth limiting material in the case
of each soil form (shown in the grey
rectangle)
QQ The area of each of the terrain units that
is occupied by the different soil series,
as well as the area occupied by rock, is
estimated in terms of percentages of the
area of that unit, as well as in hectares
(shown in the orange rectangle).
QQ The percentage and area (in hectares) of
the Land Type occupied by each of the
terrain units (numbers 1 to 5 in the terrain
form sketch; red rectangle) as well as the
slope and slope shape of each of the units,
are shown in the magenta rectangle
QQ The geology of the Land Type unit
(shown in the turquoise rectangle).
QQ The numbers of any modal profiles (see
below) in that Land Type, shown in the
brown rectangle.
Similarities between Land
Types and land systems of
engineering regional surveys
Regional surveys carried out for engineering
projects include the detailed mapping of
selected areas which are representative of
each land system. This entails the gather-
ing of general information regarding the
climate, topography, vegetation, soil and rock
comprising each selected area. Each repre-
sentative area consists of land facets such
as crests, midslopes, footslopes and valley
Volume 64 Number 1 March 2022
bottoms (Brink et al 1982). In addition, the
soils in each land facet are sampled for test-
ing for the relevant properties. Using this soil
data together with the information acquired
on the terrain of the land system ‒ such as
the approximate altitude range, approximate
relief, gradient conditions, type of surface
form and the extent of cut required in a
particular project ‒ the suitability of the land
system for the proposed development may be
evaluated (Brink et al 1982).
The Land Types and terrain units of the
Land Type Survey are comparable in infor-
mation gathered to the land systems and land
facets, respectively, of the engineering terrain
classification system (delineated during
regional surveys). The Land Types and land
systems are delineated at the same scale.
The in-depth analysis of
modal soil profiles
A number of soil profiles, termed modal pro-
files, were identified in each area covered by
a 1:250 000 Land Type sheet. The locations
of these profiles, which are indicated on the
relevant Land Type map, are such that the
range of soils occurring on the Land Type
sheet is represented by these profiles (LTSS
2021). These modal profile numbers are also
included in the Land Type inventories.
Approximately 2 500 modal profiles
were described countrywide (Patterson et al
2015). The soil occurring in the uppermost
1.2 m of each modal profile is described
according to pedological terminology and
Journal of the South African Institution of Civil Engineering
Figure 3 Example of analytical data of soils of the modal profiles (LTSS 2021)

classified according to the South African
Binomial System (MacVicar et al 1977).
An example of a modal profile description
(P137) is given in Figure 2 (LTSS 2021).
As is evident from Figure 2, substantial
information is included in the profile
description. This includes the pedologi-
cal classification, coordinates, elevation,
climate zone, terrain unit type, slope,
slope shape and details on the water
table. In addition, each layer is described
according to moisture, colour, consistency,
structure and soil type. Although some
of the standard terminology may differ
from that of Jennings et al (1973), which
is employed by the geotechnical fraternity,
the engineering terminology can be
inferred from the pedological terminology.
Furthermore, unlike the Jennings et al
(1973) method, where the origin is given
for each layer, the pedological system
states the parent and underlying materials
for the entire profile.
The quantitative physical and chemical
properties, as well as the mineralogical
composition of the soil comprising each
horizon occurring in the uppermost 1.2 m
of each modal profile, are given in the Land
Type memoirs. A typical example of the
physical, chemical and mineralogical data
given for each soil of the modal profiles is
shown in Figure 3 (LTSS 2021).
With reference to Figure 3, the follow-
ing data of particular relevance to this
paper is included:
QQ The pedological classification.
QQ The depth range of each horizon (shown
in the brown rectangle).
Table 1 Pedological and engineering soil separate sizes
Soil separate
Gravel
Coarse sand
Medium sand
Fine sand 0.2*
QQ The pedological size soil separates
(shown in the red rectangle). The sand,
silt and clay separates are expressed as
a percentage of the mass of the fraction
finer than 2 mm.
QQ The exchangeable magnesium cations,
in milli-equivalents per kg (me% × 10)
(shown in the green rectangle).
The pedological soil separate sizes, which
differ from the MIT Engineering sizes are
compared in Table 1.
A comprehensive discussion, including
typical examples, on the information that
can be obtained from the Land Type maps
and their accompanying memoirs, as well
Pedological sizes Engineering sizes (MIT)
> 2 mm > 2 mm
2 mm to 0.5 mm 2 mm to 0.6 mm
0.5 mm to 0.2* mm 0.6 mm to 0.2 mm
mm to 0.02 mm 0.2 mm to 0.06 mm

Analytical data of modal Silt 0.02 mm to 0.002 mm 0.06 mm to 0.002 mm

profile horizons
Clay < 0.002 mm < 0.002 mm
Approximately 10 000 representative soil
samples were taken from modal profiles for * The actual size used to separate the 0.2 mm size fraction in the analyses conducted for the Land
Type survey is 0.212 mm.
detailed analysis (Patterson et al 2015).

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 15
 Leslie 42 km

Ermelo 69 km
Nthorwane 33 km

Roberts Drift 1.5 km Volksrust 81 km

Roberts Drift/Standerton Road Scale 1:250 000

One or more of: vertic, melanic, red structured

Plinthic Catena: upland duplex and margalitic soils rare
Ea diagnostic horizns Bb Dystrophic and/or mesotrophic; red soils not widespread
Undefferentiated

Figure 4 Locality of the Roberts Drift–Standerton Road in relation to the Land Types

as reference to test methods, is included in investigations was illustrated by consider-
each Land Type memoir. ing an area traversed by an existing road in
the Standerton District in the Mpumalanga
Province. An approximately 21 km portion
METHODOLOGY FOR of the 22.3 km route, which extends from
UTILISING LAND TYPE DATA Roberts Drift to Standerton (Road number
FOR RECONNAISSANCE P53/3), was considered.
ENGINEERING SURVEYS The aim of this theoretical exercise was
to assess the extent to which any of the
Approach data included in the reconnaissance report,
The extent to which Land Type data could compiled prior to the construction of this
be used for reconnaissance geotechnical road, could have been determined from the
Land Type data available for this area. In
particular, emphasis was placed on the ter-
rain and soil data aspects. This was done in
the following two stages:
Stage one ‒ comprised inferring terrain
and soil information, which is relevant
for geotechnical investigations, from the
relevant Land Type inventories.
Stage two ‒ comprised determin-
ing quantitative engineering properties,
including grading, plasticity and the USCS
(USAEWES 1960) and the AASHTO system

16 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
(HRB 1945) classification groups, from the
quantitative analytical data of soils of the
modal profiles, using the model discussed in
Fanourakis (2022). This was carried out spe-
cifically for the soils previously researched
(of the Arcadia, Hutton, Shortlands,
Swartland and Valsrivier forms).
Although the model developed by
Fanourakis (2022) was based on data of soils
in an area located approximately 320 km
northeast of the Standerton area, it was
applied to the area traversed by the road, as
the soils in both areas were expected to be
similar (Harmse & Hattingh 1985).
The proposed approach was verified for
accuracy by considering the Land Types
traversed by an approximately 21 km sec-
tion of the Standerton–Roberts Drift Road
and comparing the information inferred
from the Land Type data to the actual
properties from the geotechnical investiga-
tion report for the same road.
Data used
Land Type data
Approximately 21 km of the road con-
sidered is located in the area covered by
the 2628 East Rand Land Type map and
accompanying memoir (LTSS 2021). The
remaining 1.5 km portion on the Roberts
Drift end of the road occurs in the area
covered by the 2728 Frankfort Land Type
map. The section of the road covered by
the 2628 East Rand Land Type map was
considered. Figure 4 shows the locality of
the road considered in relation to the Land
Types of the Standerton District, from the
2628 East Rand Land Type map.
Table 2 D
 ifferences between the various Land Types occurring along the Roberts Drift–
Standerton Road
Chainage (km) Land Type
1.5 – 3.4
3.4 – 10.0
10.0 – 16.0
16.0 – 17.0
17.0 – 22.3 (end)
The section of road considered tra-
verses five Land Type units. With respect
to possible Land Type boundary types on
Land Type maps, solid black indicates a
pedosystem boundary, which means that
the climate on either side is the same but the
terrain and/or soil differ on either side. Solid
red lines denote a climate and a pedosystem
boundary, which means that either climate,
terrain and soil are different on either side,
or climate and terrain are different on
either side, or climate and soil are different
on either side. Broken red lines indicate a
climate boundary only (LTSS 2021). Table 2
lists the different Land Types in relation to
the road chainages. The respects (climate,
terrain or soils) in which the Land Types
differ are also listed in this table.
Referring to the Land Type inventories,
included in the East Rand memoir (LTSS
2021), four Land Type inventories apply
along this route, namely, Ea17, Ea20, Ea24
and Ea26 (see Figure 4). The Land Types
Ea26b and Ea26c are characterised by the
same soils, terrain and climate, but are
separated by another Land Type and hence
the lowercase letter following the Ea26 is
used to indicate separated occurrences of
Difference from previous Land Type
Ea26b
Ea20g Change in climate, terrain and soils
Ea26c Change in climate, terrain and soils
Ea24a Change in climate only
Ea17e Change in terrain and soils only
the same Land Type. Furthermore, Land
Type Ea24 comprises the same soils and
terrain pattern as Land Type Ea26 but is
characterised by a different climate.
Since climate was irrelevant to this
exercise the data given in the Land Type
inventory of unit Ea26 was assumed to
apply to the area covered by Land Type
Ea24 as well. The soil and terrain invento-
ries for the three Land Type units which the
road traverses (Ea26, Ea20, Ea17) are given
as Figures 1, A1 and A2, respectively (see
Appendix for Figures A1 and A2).
A total of six Land Type modal pro-
files (P130, P136, P137, P138, P141 and
P143) occurred in the units traversed
by the road. However, P136 comprised
soil of the Sterkspruit form that was not
researched by Fanourakis (2012) and hence
was excluded.
The soil form and series of the profile,
depth of horizons and the pedological soil
separates and exchangeable magnesium (in
me %) for each horizon, for each profile,
extracted from the analytical data given for
these modal profiles, are given in Table 3.
This is the information required by the
model proposed by Fanourakis (2022) for

Table 3 Extracts of the modal profile analyses

Pedological soil separates Pedological chemical data
Modal Soil form Depth Sample < 2 mm Exch Mg/
Layer CEC* ESP*
profile and series range (mm) number > 2 mm 100 g soil
CoSa MeSa FiSa Silt Clay (me %) (%)
(me %)
A1 0–280 C2339 0 5 14 40 13 24 4.0 1.1 54.5
Valsrivier
P130 B2 280–640 C2340 1 4 9 25 11 50 10.5 1.1 136.4
Lindley
B3 640–1 200+ C2341 10 4 10 27 11 47 11.1 2.0 90

Swartland A1 0–380 C4079 4 6 21 40 7 23 2.7 11.2 1.8

P137
Nyoka B2 380–730 C4080 3 7 20 30 4 39 6.2 16.4 10.4

Arcadia A1 0–600 C2558 0 4 7 18 14 52 16.1 44.0 5.5

P138
Arcadia A3 600–1 200+ C2559 1 8 10 26 20 35 17.5 31.5 6.0
Rensburg
P141 A1 0–840 C2337 64 1 4 14 23 56 13.2 27.0 4.8
Rensburg

Arcadia A1 0–340 C4084 17 9 7 22 13 50 15.6 34.1 2.3

P143
Arcadia A3 340–490 C4085 23 17 9 24 14 38 16.3 30.2 3.3
CoSa = coarse sand; MeSa = medium sand; FiSa = fine sand
*CEC = Cation Exchange Capacity; ESP = Exchangeable Sodium Percentage

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 17
estimating grading, plasticity and engineer- Table 4 General soil groups and occupancy of each Land Type
ing classifications from pedological data. Land Type unit Ea26 Ea20 Ea17
1.5–3.4 and 17–22.3
Geotechnical report for Chainage (km)
10–17
3.4–10
(end)
Roberts Drift–Standerton Road
Depth (mm)
The data used in this exercise included
Engineering soil type
the reconnaissance report for this road
(Van Gruting et al 1975). A total of 28 % of crests
1.5 1.5 1.5
(Terrain Unit 1)
soil profiles were described as part of the
engineering investigation conducted (Van Rock % of midslopes
3
(Terrain Unit 3)
Gruting et al 1975). Disturbed soil samples
% of footslopes
were taken from a number of these profiles
(Terrain Unit 4)
and tested for the determination of their
% of valley bottoms
engineering characteristics (mainly grad-
(Terrain Unit 5)
ing, plasticity and classifications). Two of
Depth (mm) 200–400 200–300 100–250
these profiles were located within the first
Engineering soil type clayey sand clayey sand clayey sand
1.5 km of the road, which is covered by the
Frankfort Land Type map, and hence were % of crests 3 4.5 1.5
Group 1
not considered in this exercise. % of midslopes 6
% of footslopes
% of valley bottoms
TRANSLATION OF PEDOLOGICAL Depth (mm) 300–600 300–600 300–900
DATA INTO ENGINEERING DATA
sandy clay;
Engineering soil type clay clay
clay
Stage one – General properties
Group 2 % of crests 4.5 9 12
% of midslopes 15 18 35
Geology
In the three Land Type inventories (Ea26, % of footslopes 3 2 9.7
Ea20 and Ea17) pertaining to the area cov- % of valley bottoms 0.5
ered by this road, the geology of these areas Depth (mm) 250–1 200+ 200–1 200+ 150–1 200+
Percentage
is given as “dolerite, sandstone, grit and clayey sand; clayey sand; clayey sand;
of total Land Engineering soil type
shale of the Ecca Group, Karoo Sequence”. Type area sandy clay sandy clay clay
Incidentally, the Karoo Sequence is super- covered by % of crests 15 9 12
Group 3
seded by the Karoo Supergroup. According various soil
groups % of midslopes 20 18 7.5
to the geological sheet covering this area
% of footslopes 3.8 2.3 0.8
(Council for Geoscience 1986), the dolerite
% of valley bottoms
occurs mainly between chainages 10 km
and 11.5 km. The geological map also indi- Depth (mm) 200–1 200 250–500
cates tertiary deposits between chainages Engineering soil type
silty sand; sand; silty
11.5 km and 14 km. clayey sand sand

Group 4 % of crests 3
Terrain % of midslopes 10 6
According to the three relevant Land Type % of footslopes 3.8 0.5
inventories, more than 80 per cent of the land % of valley bottoms
along the entire route is characterised by
Depth (mm) 300–600 400–600 600–1 000
slopes of less than eight per cent (the terrain
sandy clay; clayey sand;
type commences with an “A”). The difference Engineering soil type clay
clay clay
between the highest and lowest points in
Group 5 % of crests
the landscape is greatest in Land Type Ea26
(between chainages 1.5 km to 3.4 km and % of midslopes 7.5

10 km to 17 km), being between 90 m to % of footslopes 4.5 0.2 4.5

150 m (denoted by a local relief class “3” after % of valley bottoms 3 4 3.5
the “A” in the terrain type. i.e. “A3”). The Depth (mm) 350–1 200 300–900 300–900
typical land pattern between these chainages silty sand; clayey sand; clayey sand;
comprises mainly (50%) of mid-slope units Engineering soil type
clayey sand sandy clay sandy clay
(with slopes of 3% to 15% and slope lengths of % of crests 3 6
Group 6
500 m to 1 000 m) and (30%) flat crests (with
% of midslopes 5 9 3
slopes of 0% to 5% and slope lengths of 400 m
% of footslopes
to 800 m). As dolerite occurs in the area,
these crests may be dolerite dykes. % of valley bottoms

18 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 Land Type unit Ea26 Ea20 Ea17
with Vertic, Melanic and Red Structured
diagnostic horizons (LTSS 2021).
Depth (mm) >1200
The percentages of the various terrain
sand; silty units (e.g. crests and midslopes) which
Engineering soil type
sand
are occupied by the various soil forms
Group 7 % of crests according to the South African Binomial
% of midslopes Classification System (MacVicar et al 1977)
Percentage
of total Land % of footslopes are given in each inventory (Figures 1, A1
Type area % of valley bottoms 0.7 and A2).
covered by From this data the area of each Land
various soil Depth (mm)
Type (expressed as a percentage) which is
groups Engineering soil type occupied by rock and various Soil Groups
Stream % of crests was determined. The seven possible Soil
beds % of midslopes Groups were determined by the author,
% of footslopes grouping soil forms which have comparable
physical properties from an engineering
% of valley bottoms 1 1 1
perspective (for example colour and/or
Legend structure and soil type). The colour, struc-
Group 1 Intact soil overlying weathered rock (Glenrosa form)
Group 2 Fissured and slickensided dark alluvial soils (e.g. Arcadia form)
ture and soil type are reflected in the char-
Group 3 Fissured non-red alluvial soils overlying – weathered rock (e.g. Swartland and Mayo forms) acteristics of the various diagnostic horizons
– rock (Milkwood form) which are discussed in the pedological clas-
– unconsolidated material (e.g. Valsrivier form) sification handbook (MacVicar et al 1977).
– ferruginised material (Tabankulu form)
– lighter-coloured fissured soil (Bonheim form) The soil type (texture), according to
Group 4 Intact soil overlying structured sandy soil (e.g. Escourt, Sterkpruit and Kroonstad forms) the pedological textural triangle of the
Group 5 Dark fissured soil overlying a gleyed horizon (e.g. Rensburg and Willowbrook forms) horizon on which series determination is
Group 6 Intact soil overlying ferruginised soil or ferricrete (e.g.Westleigh, Avalon and Wasbank forms)
based, is given for the soils of the forms
Group 7 Recent sediments or stratified alluvium (e.g. Dundee form)
comprising each Land Type in the relevant
Land Type inventory. Using the informa-
tion on soil texture recorded in the Land
Stream beds (1%) Rock (2.5%) Type inventory, a point was located in the
Group 7 (0.3%) Group 1 (5%) pedological textural triangle (MacVicar et
Group 6 (9%)
al 1977). The corresponding point was then
located in the engineering textural triangle
Group 5 (8%) (Tschebotarioff 1951) which determined
the engineering textural soil type.
Table 4 contains the occupancy of each
group, horizon depths and engineering soil
type (textural class) comprising each unit
Group 4 of each Land Type occurring along the
(9%) Group 2
Roberts Drift‒Standerton Road.
(33%)
Figure 5 shows the proportion of the
various groups, as a percentage, occupying
the section of the road considered. These
percentages consider the proportion of the
road occurring in each of the three Land
Type units. These percentages were calcu-
lated on the premise that the distribution
of the various soils within each Land Type
Group 3 (31%)
was uniform.
As is evident from Figure 5, the
dominant Soil Groups 2 and 3 occupy
Figure 5 Percentage of road route occupied by various groups approximately 33% and 31% of the route,
respectively.
The typical land pattern along the Soils Furthermore, certain inferences,
remainder of the route (Land Types Ea20 According to the soil legend which is used regarding engineering properties, can
and Ea17) consists of less local relief on the Land Type maps, the Ea units com- be made for each group, on the basis of
(compared to Land Type Ea26), comprising prise dark-coloured and/or red soils, usu- physical characteristics and the soil types
gently undulating hills that are dominated ally clayey, which have a high base status (textures), for example:
(50% to 60%) by mid-slope units and 30% (low degree of leaching). More than half QQ Groups 1 and 6 may be prone to
of crests. of this Land Type is covered by soil forms settlement.

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 19
QQ Group 2 will be prone to expansiveness
and possibly consolidation settlement.
QQ Group 3 may be prone to settlement and
expansion.
QQ Groups 4 and 7 may be prone to settle-
ment or collapse settlement.
QQ Group 5 may be prone to expansion and
settlement.
Stage 2 – Estimating quantitative
engineering soil properties
Procedure
The engineering soil properties of each
horizon were estimated by applying
the mathematical model proposed in
Fanourakis (2022), using the modal profile
data included in Table 3 as input.
However, with regard to the component
of the model that estimates engineering
grading characteristics from pedological
soil separates, there were two deviations
from the model pertaining to the estima-
tion of the percentages (by mass of the total
sample) passing 0.075 mm and 0.425 mm,
referred to as P0.075 mm and P0.425 mm,
respectively. These deviations were because
the actual size used to separate the 0.2 mm
size fraction in the analyses conducted
for the Land Type survey was 0.212 mm,
whereas in the pedological sieve analyses
conducted by Fanourakis (1999) the sieve
size was 0.25. This was appropriately
accounted for in this paper.
The engineering soil properties were
estimated by applying the mathematical
model to the modal profile data pertaining
to a given unit. A profile in the same unit
described as part of the engineering inves-
tigation (Van Gruting et al 1975) which
had physical properties (such as colour and
structure) similar to those of the modal
profile was then selected. The engineering
profile selected had to be a profile from
which samples were taken and for which
the test results were available. The esti-
mated and actual engineering properties
of the soils comprising the various modal
profiles representing each of the three Land
Types traversed by this road are respec-
tively given in Tables 5 to 7. The estimated
and actual properties for the three units
are discussed below.
Land Type Ea26 (chainages 1.5 km
to 3.4 km and 10 km to 17 km)
Only one modal profile (P141) occurred in
this unit. The soil comprising this profile
was classified as a Rensburg which con-
sists of a Vertic A horizon underlain by a
20
Table 5 Estimated and actual engineering properties of the soils occurring in Land Type Ea26
Pedological data
Modal Profile P141
Engineering properties Rensburg 20
Estimated properties
Sample number C2337
P2.0 mm 36
P0.425 mm 35
P0.075 mm 32
P0.002 mm 21
GM 1.97
LL 62
PI 33
LS 16.1
USCS SC
AASHTO A-2-7
G horizon which is calcareous in its upper
section (MacVicar et al 1977). Although
soils of the Rensburg form were not
researched in this project, the soils of the
Vertic A horizon, which solely comprise
the Arcadia form, were researched.
Eleven soil profiles were described
between the above chainages during the
reconnaissance survey (Van Gruting et al
1975). Only one of seven profiles (pro-
file SP10), from which one or more samples
were taken for testing, had physical charac-
teristics resembling those of the Rensburg
soil form.
The estimated results of the Vertic A
horizon of the modal profile were com-
pared to the actual results of sample
number SR23 from soil profile SP10 (Van
Gruting et al 1975). The estimated and
actual engineering properties of the soils in
Land Type Ea26 are given in Table 5.
Referring to Table 5, a paired t-test
conducted on the grading and plasticity
data of the mean of the estimated values
and the actual values indicated that the
difference between the two paired sets is
not significant (P = 30%). However, despite
this statistical outcome, which indicates that
the outcome of significance of difference
was not established, and not necessarily
complete acceptance of the null hypothesis,
it is evident that, practically, the estimated
and actual grading characteristics of this
horizon differ considerably. The difference
between the estimated and actual plasticity
characteristics appears to be due to the
difference in the grading characteristics
of these two soils, since the plasticity
characteristics were determined from the
Volume 64 Number 1 March 2022
Engineering data
Engineering Profile SP 10
Actual properties
Sample number
SR 23
100
97
64
14
0.39
40
20
10.5
CL
A-6
grading characteristics and exchangeable
magnesium of the soil using the appropriate
regression equation from Fanourakis (2012).
Hence, if the soil comprising the Vertic A
horizon of the modal profile had similar
grading characteristics to those of the
Vertic A horizon in profile SP10, the same
plasticity characteristics could probably be
expected from both these soils. However,
this could not be proved, as using the actual
engineering grading with the chemical data
(from modal profile P141) yielded values that
were outside of the range on which the rela-
tionships in Fanourakis (2012) were based. It
should be noted that the two soils compared
were located approximately 29 km apart
and changes in grading characteristics over
this distance are obviously not unusual.
Nevertheless, the soil type in the modal
profile description of P141 indicates this
soil to be a dry, black, very hard, blocky clay,
which would render it unsuitable for found-
ing of a road or as a road building material.
Finally, with reference to Table 4, this soil is
characterised as Group 2, which confirms
its unsuitability.
Land Type Ea20 (chainage
3.4 km to 10 km)
A total of two modal profiles occurred
in this unit (P130 and P137), as shown in
Table 3. Modal profiles P130 and P137
comprised soils of the Swartland and
Valsrivier forms, respectively (LTSS 2021).
Eleven profiles were recorded between
the above chainages as part of the engi-
neering reconnaissance survey conducted
(Van Gruting et al 1975). These profiles
comprised soils of the Swartland, Arcadia,
Journal of the South African Institution of Civil Engineering
Table 6 Estimated and actual engineering properties of the soils occurring in Land Type Ea20 by Fanourakis (2012), which were 2.2 to
Engineering 4.4 milli-equivalents per 100 g soil.
Pedological data QQ The exchangeable sodium percentages
data
(ESPs) were in excess of 50% (ranging
Modal Profile P137 Mean of Engineering
Engineering Swartland 41 estimated from 55% to 136%), whereas those of
Profile SP 13
properties Estimated properties properties Actual the soils researched, and on which the
Sample properties relationships in Fanourakis (2012) were
Sample number Sample number Sample number
numbers based, were generally below 5%.
C4079 C4080 SR 29
C4079 & C4080 The mathematical model was then applied
P2.0 mm 96.0 97.0 97 100 using the pedological data from profile
P137. The estimated and actual engineering
P0.425 mm 86.1 86.5 86 98
properties of the soils in Land Type Ea20
P0.075 mm 51.9 58.0 55 60 are given in Table 6.
P0.002 mm 22.8 37.8 30 16 Referring to Table 6, a paired t-test con-
ducted on the grading and plasticity data of
GM 0.66 0.58 0.60 0.42
the mean of the estimated values and the
LL 30.1 38.6 34 33 actual values indicated that the difference
between the two paired sets was not sig-
PI 12.6 17.1 15 16
nificant (P = 78%). Hence the null hypoth-
LS 6.5 9.2 7.9 8 esis was not rejected. Furthermore, the
USCS CL CL CL CL Atterberg Limits in the two sets compared
were almost identical and the classification
AASHTO A-6 A-6 A-6 A-6
groups are identical.

Table 7 Estimated and actual engineering properties of the soils occurring in Land Type Ea17 Land Type Ea17 (chainage 17 km to end)
Engineering
Two modal soil profiles occurred in this
Pedological data unit, P138 and P143 (LTSS 2021). Each
data
of these profiles comprised two Vertic A
Modal Profile P138 Modal Profile P143 Engineering
Engineering Arcadia 40 Arcadia 40 Profile SP 17 horizons and are classified as Arcadia
properties Estimated properties Estimated properties Mean of Actual according to the South African Binomial
estimated properties Classification System (MacVicar et al
Sample Sample Sample Sample properties Sample
number number number number number
1977). Five soil profiles were recorded
C2558 C2559 C4084 C4085 SR 32 between the above chainages during the
reconnaissance surveys (Van Gruting et
P2.0 mm 100 99 83.0 77.0 90 99
al 1975). Of these five profiles, profile
P0.425 mm 94.4 89.1 74.5 62.9 80 95 SP17 comprised soil which had the physi-
cal characteristics of a Vertic A horizon.
P0.075 mm 80.1 69.6 61.9 49.4 65 74
Sample number SR32, taken from this hori-
P0.002 mm 54.7 35.0 41.1 28.7 40 7 zon which appeared to have the physical
GM 0.26 0.42 0.81 1.11 0.6 0.32 properties of a Vertic A horizon, was tested
to determine its engineering properties
LL 64.6 60.4 63.4 61.3 62 64
(Van Gruting et al 1975).
PI 34.0 31.7 33.3 32.2 33 36 The model was applied to the four sets
of pedological data pertaining to the two
LS 16.6 15.8 16.4 16.0 16 14
horizons of each of the two modal profiles
USCS CH CH CH SC** CH CH to determine their engineering properties.
The estimated and actual engineering
AASHTO A-7-5 A-7-6 A-7-5* A-7-6 A-7-6 A-7-6
properties of the soils in Land Type Ea17
* Borderline: If PI was 0.2 higher, classification would be A-7-6. are given in Table 7.
** Borderline: If P0.075 mm was 0.1 higher, classification would be CH.
Referring to Table 7, a paired t-test
conducted on the grading and plasticity
Rensburg, Katspruit, Glenrosa and Mispah The plasticity estimation component of data of the mean of the estimated values
forms. This was determined by simply the model could not be applied using the and the actual values indicated that the dif-
inspecting the physical characteristics of modal profile data from profile P130 for the ference between the two paired sets is not
the horizons in each of the profiles. Only following reasons: significant (P = 95%). This non-rejection
one of the profiles from which samples QQ The cation exchange capacities (CECs) of of the null hypothesis was despite the dif-
were taken for testing (SP13) had the physi- the soils comprising the three horizons ference in clay content (40% and 7%) of the
cal characteristics of the Swartland form. in this profile ranged from 1.1 to 2 milli- two groups compared. The actual reported
Sample number SR29 was tested from this equivalents per 100 g soil. These were clay content value of 7% (in the road report)
profile (Van Gruting et al 1975). lower than those of the soils researched is questionable, as this indicates an activity

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 21
of clay in excess of five. Although such high
activities of clay are possible (Skempton
1953), the similar soils of the Arcadia form
researched comprised activities of clay
ranging from 0.47 to 1.95. Finally, in the
two sets compared, the Atterberg Limits
were almost identical, and the classification
groups were identical.
General comparison of the
road reconnaissance report
to Land Type data
In the reconnaissance report the entire
area was mapped as a single land pattern
or unit, whereas according to the Land
Type map the road is located across three
different units. The section of the recon-
naissance report dealing with the soils
along the proposed road briefly discusses
the dominant alluvial soils and colluvial/
residual soils separately.
The alluvial soils have a sandy to clayey
silt texture and are present over most of the
area mapped. These soils extend to depths
between 1.0 m and 4.0 m, and are thought
to have been deposited in the past by the
Vaal River. Mention is also made that
certain zones which may contain expansive
or heaving clays were identified during the
aerial photographic interpretation (Van
Gruting et al 1975).
The surface soils which are colluvial or
residual are scattered throughout the area
and are only associated with the plateaux
and raised dolerite ridges. The colluvial
soils, which extend to depths ranging
from 0.1 m to in excess of 1.0 m and have
a silty to sandy clay texture, are usually
underlain by residual dolerite or shale along
the side slopes of the ridges (Van Gruting
et al 1975).
The discussion on the soils in the
reconnaissance report essentially contains
the same information as data given in the
Land Type inventories of the Land Types
traversed by the road.
The engineering soil properties estimat-
ed from the quantitative modal soil profile
data were all statistically significant and
compared favourably to those of the same
soil types tested during the reconnaissance
survey, in the case of 86% (six of the seven)
sets of data. The differences in the grading
characteristics of the soils compared may
be attributed to the horizontal distances
(ranging from 29 km to 5.6 km) between
the modal profiles and the profiles to
which they were compared.
No modal profiles were recorded in the
dolerite areas. This is of great importance
22
when establishing potential borrow areas.
In addition, the probability that a modal
profile will be located on a route which in
future may be used for the construction
of a road or in a potential borrow area is
slight. In general, when investigating bor-
row pits, it is the pedocretes (such as pedo-
logical “C”-horizons, etc) or weathered rock
which are of interest and warrant testing.
In the pedological soil classification system
none of the soil series are determined
on the properties of the pedocretes or
weathered rocks, and hence no indication
of the physical and/or chemical composi-
tion of the pedocretes is given in the soil
series. However, the pedological profile
may indicate that pedocretes do occur in a
particular area.
The geological data supplied in the
Land Type inventories are merely tran-
scriptions of the standard geological survey
data which appear on the geological maps
(Council for Geoscience 1986) and are,
therefore, brief. The geological data in the
engineering reconnaissance report is given
in detail, including a geological cross-
section of the route.
CONCLUSIONS
Land Type data, together with the model
developed, may be successfully used, ide-
ally by appropriately qualified registered
geotechnical professionals, to plan recon-
naissance surveys. Soils of the same soil
form with physical and chemical proper-
ties as the soils included in the research
project, but which occurred outside the
area in which the research was conducted,
displayed statistically significant (P = 30%
to P = 95%) estimated engineering proper-
ties (when compared to the actual proper-
ties). In 86% of the data sets the estimated
properties were almost identical to the
actual properties.
This case study highlighted the fact
that data of engineering relevance deter-
mined using the Land Type memoirs and
the mathematical model by Fanourakis
(2022) may be used to obtain an indica-
tion of materials, drainage characteristics,
subgrade conditions and potential problem
soils occurring in an area or along a route.
Such information would be of benefit,
including during reconnaissance geotech-
nical surveys, and in planning and costing
detailed investigations.
Since the information contained in
the Land Type maps is (because of the
small scale used) too general to enable
Volume 64 Number 1 March 2022
the location of borrow material, it appears
that the use of conventional aerial pho-
tographic interpretation techniques is a
more effective method of locating potential
borrow areas.
References
Brink, A B A, Partridge, T C & Williams, A A B 1982.
Soil Survey for Engineering. New York: Oxford
University Press, pp 225‒227, 258.
Calitz, F 2021. Proposed system to facilitate use
of pedological information in preliminary
stage geotechnical investigations. PhD Thesis.
Potchefstroom: North-West University.
Council for Geoscience 1986. Geological Series
Map 2628 East Rand. Scale 1:250 000. Pretoria:
Government Printer.
Fanourakis, G C 1999. Engineering soil properties from
pedological data. DTech (Eng) Thesis. University of
Johannesburg.
Fanourakis, G C 2012. Estimating soil plasticity
properties from pedological data. Journal of the
South African Institution of Civil Engineering,
54(2): 117‒125.
Fanourakis, G C 2022. A mathematical model for
determining engineering soil classifications from
pedological data. Journal of the South African
Institution of Civil Engineering, 64(1): 2‒11.
Jennings, J E, Brink, A B A & Williams, A A B 1973.
Revised guide to soil profiling for civil engineering
purposes in Southern Africa. Transactions of
the South African Institution of Civil Engineers,
15(1): 3‒12.
Harmse, H J Von M & Hattingh, J M 1985. The
pedological profile classes. In Brink, A B A (Ed.),
Engineering Geology of Southern Africa, Vol 4,
272–273.
HRB (Highway Research Board) 1945. The United
States Bureau of Public Roads Classification System.
Proceedings, 25th Annual Meeting of the HRB.
Washington, DC.
ISRIC (International Soil Reference and Information
Centre) 2021. World soil information: Application of
soil information. Wageningen, Netherlands: ISRIC.
www.isric.org/explore/library.
LTSS (Land Type Survey Staff) 2021. Land types of
South Africa: Digital Map (1:250 000 scale) and
soil inventory databases. Pretoria: Institute for Soil,
Climate and Water, Agricultural Research Council.
https://www.arc.agric.za/arc-iscw/Pages/ARC-
ISCW-Homepage.aspx.
MacVicar, C N, De Villiers, J M, Loxton, R F, Verster, E,
Lambrechts, J J N, Merryweather et al 1977. Soil
classification: A binomial system for South Africa.
Scientific Bulletin No 390. Pretoria: Department of
Agriculture Technical Services.
Paige-Green, P & Turner, D 2007. The use of soil and
land type information maps in road engineering in
South Africa. Proceedings, 14th African Regional
Journal of the South African Institution of Civil Engineering
 Conference on Soil Mechanics and Geotechnical
Engineering (ARC SMGE), Yaounde, Cameroon,
26–28 November 2007, pp 311‒318.
Patterson, G, Turner, D, Wiese, L, Van Zijl, G, Clarke, C
& Van Tol, J 2015. Spatial soil information in South
Africa: Situation analysis, limitations and challenges.
South African Journal of Science, 111(5/6): 1‒7.
Skempton, A W 1953. The colloidal activities of clays.
Proceedings, 3rd International Conference on Soil
Appendix
Figure A.1 Land Type inventory of Ea20 (LTSS 2021)
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
Mechanics and Foundation Engineering, Zurich,
Switzerland, pp 57‒61.
Tschebotarioff, G P 1951. Soil Mechanics, Foundations
and Earth Structures. New York: McGraw-Hill.
USAEWES (United States Army Engineer Waterways
Experiment Station) 1960. The Unified Soil
Classification System. Technical Memorandum
No 3-357. Appendix A, Characteristics of soil groups
pertaining to embankments and foundations, 1953;
Appendix B, Characteristics of soil groups pertaining
to roads and airfields. Vicksburg, MS: USAEWES.
Van Gruting, Comninos & Potter 1975. Reconnaissance
report on soil engineering mapping, terrain
evaluation and construction materials location
for the proposed Standerton – Roberts Drift Road
(P53/3). Undertaken by P G Hartopp and Associates
(Johannesburg) and B W Kraft (Pretoria).
 Appendix continued 
23
Figure A.2 Land Type inventory of Ea17 (LTSS 2021)

24 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Incorporation of additional TECHNICAL PAPER
information into the Journal of the South African
Institution of Civil Engineering

South African Wind ISSN 1021-2019

Vol 64 No 1, March 2022, Pages 25–36, Paper 1266

Load Formulation DR FREDERIK BAKKER obtained a BEng

(Civil) degree in 2017 and a PhD (Structural
Engineering) in 2021, both from
Stellenbosch University. His research
F P Bakker, N de Koker, C Viljoen interests are in structural reliability, wind
loading, and informatics. He is currently
employed as a Software Engineer at Trimble
Quest in Cape Town.
The South African wind loading standard SANS 10160-3:2019 recently adopted an improved
map of characteristic basic wind speeds and increased the wind loading partial safety factor Contact details:
Department of Civil Engineering
from 1.3 to 1.6. These changes represent an overhaul of the design wind loads throughout Stellenbosch University
South Africa and were the result of several studies on the wind loading standard. Since these Private Bag X1
studies were conducted, substantially more wind speed data has been made available. This Matieland 7602
South Africa
investigation aimed to use this data to assess the current design loads by estimating location-
T: +27 84 392 1123
specific design values that maintain the current reliability level of the standard. A statistical E: fred_bakker@trimble.com
test was developed to assess whether the design values in SANS 10160-3:2019 could be
supported by the new data. It was found that several updates could be considered. These were PROF NICO DE KOKER (Pr Eng, MSAICE) is a
incorporated into a new recommended map of basic wind speeds that could be considered for computational physicist and structural
engineer, currently serving as Head of the
inclusion in the next revision of SANS 10160-3.
Department of Civil Engineering at
Stellenbosch University. Following many
years specialising in thermodynamics and
NOMENCLATURE cki characteristic value of a specific heat transfer in materials at extreme
conditions, he shifted his focus to
component of the load modifi-
engineering, completing a second PhD in reliability-based design through
Wind load cation factor Stellenbosch University. His research interests focus on the analysis of
q random variable representing cprob factor used to adjust the char- uncertainty in engineering calculations, and its expression in design via
the wind loading acteristic value for different risk and reliability.

qd design wind load design lives Contact details:

qd:sans current SANS design wind load cprob:asce factor to adjust the characteris- Department of Civil Engineering
Stellenbosch University
q̂d estimate of design wind load tic value to different design lives Private Bag X1
γ w wind load partial safety factor in ASCE Matieland 7602
cprob:sans factor to adjust the characteris- South Africa
T: +27 21 808 4434
Wind speed tic value to different design lives
E: ndekoker@sun.ac.za
v wind speed in SANS
v50 50-year characteristic value of PROF CELESTE VILJOEN (Pr Eng, FSAICE) is
the basic wind speed Probability currently Vice-Dean Teaching in the Faculty
v50:updated updated 50-year characteristic P probability of Engineering at Stellenbosch University,
after previously serving as Head of the
wind speed p annual exceedance probability
Division of Structural Engineering and Civil
vd design wind speed pt target annual exceedance Engineering Informatics at the same
v̂d estimate of design wind speed probability university. Her research considers structural
vsim simulated wind speed pt:vd annual exceedance probability risk and reliability, including the assessment
of various structural standard provisions in reinforced concrete design,
vw preconditioned wind speed of the design wind speed
steel design, and loading. She has been widely involved in structural
w wind speed preconditioning βt target reliability index standardisation, notably as the convenor of the SANS 10100-3 Working
exponent c FORM sensitivity factor of the Group, member of the Working Group for the Revision of ISO 13824, and
wind load modification factor member of the International Joint Committee on Structural Safety.

Wind load modification factor αv FORM sensitivity factor of the Contact details:
c wind load modification factor wind speed Department of Civil Engineering
Stellenbosch University
ck characteristic wind load modifi- Φ standard cumulative normal Private Bag X1
cation factor distribution function Matieland 7602
cd design value of the wind load θ standard score, i.e. number of South Africa
T: +27 21 808 4444
modification factor standard deviations away from
E: celesteviljoen@sun.ac.za
ci a specific component of the load the mean
modification factor, e.g. pres- θt standard score of the design
sure coefficient value
Keywords: wind loading, structural reliability, wind engineering

Bakker FP, De Koker N, Viljoen C. Incorporation of additional information into the South African Wind Load Formulation.
J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #1266, 12 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64no1a3 25
 θk:opt optimal standard score of the
characteristic value
Wind speed statistics
δvw coefficient of variation of pre-
conditioned wind speeds
δv 2 coefficient of variation of
squared wind speeds
δv1.6 coefficient of variation of wind
speeds raised by an exponent of
1.6
δv1.6 average coefficient of variation
of wind speeds raised by an
exponent of 1.6
δ̂ v1.6 estimate of mean coefficient of
variation of wind speeds raised
by an exponent of 1.6
δv 2:syn average coefficient of variation of
synoptic-squared wind speeds
δv 2:ts average coefficient of variation
of thunderstorm-squared wind
speeds
μvw mean of the preconditioned
wind speed
μ̂v1.6 sample estimates of mean of
wind speeds raised by an expo-
nent of 1.6
μ̂v1.6:sim sample mean of simulated wind
speeds raised by an exponent of
1.6
σ vw standard deviation of the pre-
conditioned wind speed
σ̂v1.6 sample estimates of standard
deviation of wind speeds raised
by an exponent of 1.6
σ̂v1.6:sim sample standard deviation of
simulated wind speeds raised by
an exponent of 1.6
λ skewness
κ kurtosis
Θ set of distribution parameters
Θsim simulated set of distribution
parameters
Other
ρ air density
l limit state
r random variable representing
the structural resistance
rd design value of structural
resistance
g random variable representing
the permanent loading
gd design values of permanent
loading
n, M, m some number of
f a function of
∆ sans relative difference between the
predicted and the current SANS
design wind load
26
γ Euler’s gamma = 0.577…
α* statistical significance level
H0 null hypothesis
r proportion of simulated design
values that are less than the
current SANS design load
INTRODUCTION
The current South African wind loading
standard SANS 10160-3:2019 (SANS 2019b)
is largely based on the Eurocode EN 1991‑1‑4
(EN 2005), and uses Davenport (1982)’s
wind loading chain to convert a representa-
tive wind speed v50 to a design wind load/
dynamic pressure as
1 In SANS 10160-3:2019, one of four values
qd:sans = ργ c v 2 (1)
2 w k 50
m
ck = Π cki
i=1
where ρ is the air density. The v50 value
is the characteristic value of the basic
wind speed, with an annual exceedance
probability p = 0.02, representing wind
speed measured at 10 metres above ground
level in open country terrain with low
vegetation. The load is tailored to site and
structural conditions using ck, which is the
product of other factors cki. These account
for different aspects of the particular
load, including the change in load due to
height, topography and aerodynamics of
the structure. To ensure that the calculated
load meets the reliability requirements of
Figure 1 Characteristic wind speed map in SANS 10160-3:2019
Volume 64 Number 1 March 2022
the standard SANS 10160-1:2019 (SANS
2019a), uncertainties are accounted for
using the partial safety factor γ w .
New wind speed data has recently been
made available (Bakker & Viljoen 2019) and
more research into applicable statistical
techniques has been conducted (Bakker
et al 2021; Bakker 2021), which present an
opportunity to improve the representa-
tion of the South African wind climate in
SANS 10160-3.
Basic wind speed
In Equation 1 the wind climate is explic-
itly represented by the basic wind speed
v50. Different v50 values are specified to
account for regional differences in climate.
v50 ∈ {32, 36, 40, 44} m/s is specified for
each district and metropolitan municipal-
ity (administrative divisions of a South
African province) using a characteristic
wind speed map (Figure 1) and table. SANS
10160-3:2019 introduced this map, which
represents a substantial improvement
over previous maps. Before Figure 1 was
adopted, the characteristic map was based
on studies which only had access to sparse
data (Milford 1985; 1987). A background
on the various basic wind speed maps that
have been used in South Africa since 1952
is presented by Goliger et al (2017).
The current map originates from an
analysis of gust wind speeds collected at 92
weather stations throughout South Africa
performed by Kruger (2011), who estimated
Wind speed
32 m/s
36 m/s
40 m/s
44 m/s
Journal of the South African Institution of Civil Engineering

Figure 2 Distribution and dominant climatic mechanism of stations used in this study
the annual p = 0.02 exceedance gust wind
speeds using Peak-Over-Threshold (POT)
and Gumbel analyses (Palutikof et al
1999). The derivation of the current map is
described by Kruger et al (2017).
Cook (1985) recommends at least 20
years of data for reliable Gumbel results,
but data used by Kruger (2011) was limited
to the period between 1990 and 2008, with
a longest series of only 16 years, which
is why Kruger (2011) also performed the
POT analysis. To be conservative, Kruger
(2011) then adopted the maximum 75th
percentile gust wind speed of the Gumbel
and POT analysis.
In response to the shortage of data, an
effort to increase the quantity of available
quality data has been made (Bakker &
Viljoen 2019). This yielded an extended
dataset which includes over 3 500 annual
gust maxima from 132 stations classified
by dominant storm type and normalised
to standardised conditions. This is more
than a twofold increase in data compared
to Kruger (2011)’s study. The location and
dominating extreme wind speed mecha-
nism of each station are shown in Figure 2.
Bakker et al (2021) considered the
extended dataset from Bakker and Viljoen
(2019) in terms of model selection and found
that a Gumbel analysis was preferred. The
downside of using the Gumbel distribution
is that its inflexibility relative to distribu-
tions with more parameters, such as the
Generalised Extreme Value distribution,
could introduce significant modelling
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
Stations
Extratropical
cyclones
Mixed climate
Thunderstorms
bias. To reduce modelling bias without a
significant increase in statistical uncertainty,
Bakker et al (2021) recommended regional
preconditioning of the wind speed (raising
the wind speeds by an exponent w) before
fitting the Gumbel distribution. Bakker et
al (2021) developed a maximum likelihood-
based procedure to estimate this exponent
and found that a value of w = 1.6 should
lead to a reduction of modelling bias, given
the extended South African dataset. This
technique and the new data could be used
to update the v50 values in SANS 10160-3,
which should improve its representation of
the extreme wind climate.
Reliability performance
In Equation 1 the wind climate is also
implicitly represented by the partial factor
γ w . The γ w accounts for uncertainties in the
load, a significant portion of which can be
attributed to the wind climate (Ellingwood
& Tekie 1999; Holický 2009; Botha et al
2018b). A reliability assessment by Botha et
al (2018a) led to the γ w = 1.6 currently used
in SANS 10160-1:2019 (SANS 2019a).
A reliability assessment aims to ensure
that a structure designed according to the
standard has an acceptably low lifetime
probability of failure. This requirement can
be considered for a wind loading standard
using a limit state l of
l = r – g – q(2)
q = cv 2
where r, g and q are random variables
representing the structural resistance,
permanent loading, and wind loading
respectively. The wind loading can be
decomposed into two separate random
variables: the wind load modification factor
c and the wind speed v.
An acceptable design is found by
specifying a set of design values {rd, gd,
qd} which ensure that the probability of
a load exceeding the resistance P(l < 0)
equals some target probability pt. This pt
is associated with a specified target level of
reliability and is typically referred to using
the target reliability index βt = Φ–1(1 – pt),
with Φ the standard cumulative normal
distribution function. Because structures
differ in performance, consequence of
failure, and nature of failure, different βt
values are specified for different reliability
classes (Retief & Dunaiski 2009). Deviation
from these βt values should be minimised
throughout the scope of the standard.
If distributions of the random vari-
ables can be defined, then a set of optimal
design values, at the most probable point
of failure, can be found. An efficient way to
achieve this is provided by the First Order
Reliability Method (FORM) (Ang & Tang
1984). This also provides sensitivity factors
α which indicate the relative contribution
of each component.
Botha et al (2018a) considered a single
representative distribution of dynamic wind
pressure (equivalent to v2) and a range of
g, r and c distributions in their reliability
assessment of SANS 10160-3. Unlike the
probability distribution of r, g and c, which
can reasonably be considered location
independent, the distribution of v could
differ by climatic region and thus vary by
location (Ellingwood & Tekie 1999). This
implies that using a single distribution of v
to derive a single γ w may lead to geographi-
cal variation in reliability performance. This
may be acceptable if the region covered by
the standard has a fairly homogeneous wind
climate (Hong et al 2016), as is the case for
many countries that use EN 1991-1-4.
Larger regions are likely to experience
substantial variation in the wind climate
and could see significant geographical
variation in reliability performance if a
single γ w is specified for the entire region.
Therefore, standards that aim to cover
diverse climatic regions, including the
American ASCE 7-16 (ASCE 2017)) and
Australian AS/NZS 1170.2 (AS/NZS 2011)
standards, adapted the partial factor-based
format of Equation 1.
27
 Instead, a high return period design
wind speed vd that corresponds with the
target reliability is specified, effectively
setting γ w = 1. This ensures that, as long as
regionally representative statistics are used
to estimate vd, the regional variation in the
wind climate is accounted for in the design
(Hong et al 2016). The strong wind climate
throughout South Africa is diverse (Kruger
et al 2010), and so an approach similar to
ASCE 7-16 (ASCE 2017) and AS/NZS 1170.2
(AS/NZS 2011) could be considered to yield
more uniform reliability performance.
ESTIMATION OF DESIGN
WIND SPEEDS
Direct consideration of
design wind speeds
The use of a single partial factor across an
area with some regions dominated by syn-
optic storms and others by tropical cyclones
is inappropriate because the variability
of wind speeds from tropical cyclones is
typically higher than synoptic storms. This
has been recognised in standards that need
to cover both cyclone and synoptic storm
regions; in the United States and Australia
an “importance factor” / “cyclone factor”
was used to increase design loads to account
for the higher variability of tropical cyclones
(Holmes 2018). In both standards this prac-
tice has been superseded and a design wind
speed vd is used instead. The exceedance
probability of this design wind speed pt:vd is
specified to correspond to the target reliabil-
ity. This avoids the need to have different
regional partial factors because the regional
differences are reflected in vd.
ASCE 7-16 (ASCE 2017) and AS/NZS
1170.2 (AS/NZS 2011) specify multiple
vd values at a given location depending
on the reliability class (risk categories in
ASCE 7-16 (ASCE 2017) or importance
levels in AS/NZS 1170.2 (AS/NZS 2011))
because the difference in variability
between cyclonic and non-cyclonic winds
is too large to reasonably allow conversion
between pt:vd values using a single formula.
South Africa has a diverse wind cli-
mate, dominated by synoptic winds in the
southwest and thunderstorm winds in the
northeast (Kruger et al 2010). These dif-
ferent climatic mechanisms were found to
have significantly (P-value <≈ 0.01) different
coefficients of variation of v 2 with means of
δv 2:syn = 0.24 and δv 2:ts = 0.29 for synoptic
and thunderstorm regions respectively
(Botha 2016).
28
Figure 3 Interpolated coefficient of variation of v 2 throughout South Africa
This trend was further investigated
using the dataset compiled by Bakker and
Viljoen (2019). The coefficient of variation
δv 2 of the 170 available series was esti-
mated, then inverse distance weighting was
used to interpolate between stations, and
δv 2 was aggregated by municipality. This
yielded a map of δv 2 as shown in Figure 3.
There is a clear increase in δv 2 from south-
west to northeast that correlates with the
dominant climatic mechanism (Figure 3),
confirming the trend observed by Botha
(2016) of synoptic-dominated regions
having a lower δv 2 and thunderstorm-
dominated regions a higher δv 2 .
Botha et al (2018a) aimed to perform a
reliability assessment for the country as a
whole and used δv2 = 0.31 to calibrate the
partial factor γ w = 1.6 for the entire country.
While this δv2 was a reasonable choice for
nationwide consideration, Figure 3 suggests
it may have been overly conservative for
synoptic regions in the southwest of the
country. Therefore, direct consideration
vd of values is proposed to obtain more
regionally representative design loads. An
advantage of directly considering vd instead
of the characteristic values (p = 0.02) is that
statistical techniques can be applied directly,
and the full impact of the new information
(Bakker & Viljoen 2019; Bakker et al 2021;
Bakker 2021) can be realised.
The vd should be estimated for an
exceedance probability pt:vd that will result
in attainment of the target reliability.
This pt:vd could be found by performing
Volume 64 Number 1 March 2022
Coefficient
of variation
0.24–0.26
0.26–0.27
0.27–0.28
0.28–0.30
0.30–0.32
a reliability assessment using FORM and
finding the pt:vd from the design point.
SANS 10160-3:2019 is meant to apply to a
wide range of design situations (Retief &
Dunaiski 2009) and so multiple resistance
distributions and wind to total load ratios
would need to be considered, which would
lead to ambiguity about which design
point to use. Alternatively, because Botha
et al (2018a) have already calibrated the
reliability level of the standard, the pt:vd
currently provided by SANS 10160-3:2019
could be calculated and used instead. This
would avoid having to repeat the subjective
decisions and calibration procedure done
by Botha et al (2018a).
ASCE exceedance probability
The pt:vd used by ASCE 7-16 (ASCE 2017)
was back-calculated from provisions in
ASCE 7-05 which used a similar format
to SANS 10160-3:2019 (Cook et al 2011)
with design wind loads also calculated
using Equation 1. If it is assumed that the
variability of the wind speed dominates the
uncertainties accounted for by the partial
factor γ w, an alternative formulation using
a design wind speed vd is:
qd = ck v 2d(3)
Combining this with Equation 1 means vd
can be represented in terms of the previous
provisions as:
v 2d = γ wv 250(4)
Journal of the South African Institution of Civil Engineering

10 –6
10 –5
10 –4
pt
10 –3
10 –2
10 –1
0.8 1.0
Figure 4 R
 elationship between exceedance probability and partial factor, using the method
described by Cook et al (2011)
ASCE 7-05 allowed adjustment of the char-
acteristic value to another annual exceed-
ance probability p using:
vk = cprobv50(5)
where cprob is a factor used to adjust the
characteristic value for different design
lives, which in ASCE 7-05 is:
cprob:asce = [0.36 + 0.1 ln(12/p)](6)
Combining this with Equation 4 gives:
⎫ 12 ⎫ (7)
√γ w = 0.36 + 0.1 ln⎪ ⎪
⎭ pt:vd ⎭
from which the pt:vd that matches the
reliability of the wind speeds can be
determined. For example, the ASCE 7-16
(ASCE 2017) reference level of reliability
(Risk Category II) with γ w = 1.6 gives
pt:vd = 1.4 × 10 –3 (700-year return period).
The similarities between ASCE 7-05
and SANS 10160-3:2019 mean that the
same procedure could be considered for
South Africa.
In SANS 10160-3:2019 the adjustment
of v50 to a different annual exceedance
probability is done with:
0.5
1 – 0.2 ln(– ln(1 – p))
cprob:sans =
1 – 0.2 ln(– ln(0.98))
The relationship between pt and γ w result-
ing from cprob is depicted in Figure 4, where
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
1.2 1.4 1.6 1.8 2.0
γw
SANS ASCE
a significant discrepancy between ASCE
and SANS is evident. The partial factor in
SANS 10160-1:2019 is also γ w = 1.6, but
cprob:sans implies an annual pt = 10 –4, which
is more conservative than ASCE by an
order of magnitude.
The ASCE method of determining
pt:vd assumes that v dominates the uncer-
tainty of the wind loading formulation.
Prescribing all the uncertainty to v means
that the variability of v dictates the design
load qd . Therefore, if uncertainties in c are
large and pt:vd is calculated neglecting these
uncertainties, then the pt:vd obtained would
be too far into the tail of the distribution,
resulting in too much weight being given
to the variability of v. Thus, qd would be
overestimated in areas where the variation
of v is high and underestimated in areas
where the variation of v is low.
Research by Botha et al (2018b) found
that uncertainties in other components
of Davenport (1982)’s wind loading chain
could be significant. These findings were
incorporated into the Botha et al (2018a)
calibration of SANS 10160-3; hence,
neglecting uncertainties in c when back-
calculating the target reliability is not
considered reasonable.
Most probable exceedance
probability
(8)
If uncertainties in c are considered, the
design wind load can be represented as:
qd = cdv 2d(9)
which can be combined with Equation 1
to give:
γw = ⎫⎪ d ⎫⎪⎫⎪ d ⎫⎪(10)
c v2
⎭ ck ⎭⎭ v 250 ⎭
Unlike the ASCE method, there are multiple
possible vd values, with corresponding cd
values, which together satisfy Equation 10.
Defining exceedance of the design load
as failure means that:
l = γ w – ⎫⎪ ⎫⎪⎫⎪ ⎫⎪(11)
c v 2
⎭ ck ⎭⎭ v 250 ⎭
represents an acceptable limit state, with c
and v as random variables. Assuming {vd, cd}
are at the design point, solving P(l < 0) using
FORM provides both the probability βq
that the wind load exceeds its design value
and the most probable {vd, cd} solution. The
exceedance probabilities of cd and vd are
then defined by their sensitivity factors, for
example pt:vd = 1 – Φ(αvβq).
The limit state can be formulated in
terms of preconditioned wind speeds (as
defined in the Introduction) vw as:
⎫ c ⎫⎫ vw ⎫
l = γ w – ⎪ ⎪⎪ ⎪
2 ∕w
⎭ ck ⎭⎭v50
w⎭
(12)
Bakker et al (2021) investigated w using the
extended dataset from Bakker and Viljoen
(2019) and found that a w ≈ 1.6 fits the data
significantly better than if the Gumbel dis-
tribution were applied to the wind speeds
(w = 1) or dynamic wind pressures (w = 2).
Applying w = 1.6, and since in South
Africa γ w = 1.6,
⎫ c ⎫⎫ v1.6 ⎫
l = 1.6 – ⎪ ⎪⎪ ⎪
2 ∕1.6
⎭ ck ⎭⎭v50
1.6⎭
(13)
The FORM P(l < 0) solution can then be
found if the distributions of c and vw rela-
tive to their values in the standard c∕ck and
v1.6 ∕v50
1.6 are defined.
Distribution of v
A Gumbel distribution is often used to
represent the distribution of wind speeds
(Botha et al 2018a; Kruger 2011; Hong et
al 2014; JCSS Model Code (JCSS 2001)). A
generalisation of the Gumbel distribution
to vw has a cumulative density function of:
1 –p = e xp –exp
(πvw – πμvw + γ√6σ vw)
���
���
– (14)
√6σ vw
29
Table 1 M
 ean and standard deviation of combined lognormal c∕ck distributions defined by Botha et al (2018a) and results of FORM analysis

Model E[c∕ck] STD[c∕ck] βq αv αc cd ∕ck pt:vd

New-SANS 0.7 0.28 3.11 0.68 0.73 1.59 16.7 × 10 –3

Updated Milford 0.57 0.18 3.83 0.77 0.64 1.16 1.7 × 10 –3

Updated Gulvanessian / Holicky΄ 0.67 0.19 3.61 0.80 0.60 1.18 2.0 × 10 –3

Updated Holicky΄ 0.84 0.19 3.32 0.85 0.52 1.20 2.3 × 10 –3

where γ = 0.577 … is Euler’s gamma. A log-normal distribution has also reference target exceedance probability
The μvw and σ vw are the mean and been used for ci ∕cki (JCSS 2001; Baravalle & βt = 4.0. The high variability of c∕ck in the
standard deviation of the preconditioned KÖhler 2018; Hong et al 2016). The product New-SANS model means that, unlike the
wind speed vw. Equation 14 can also be of a series of log-normally distributed other three models, αc > αv indicates that c
represented as: random variables also follows a log-normal contributed more to the total uncertainty
distribution, with parameters defined by its of the wind load than v. This directly con-
v = [μvw + θ(p)σ vw]1 ∕w(15) component distributions. Therefore, if each tradicts the ASCE method, where αc = 0
ci ∕cki is assumed to be log-normally dis- was assumed.
where θ is a standardisation function given tributed with mean and standard deviation The updated models all have a fairly
as: defined by Botha (2016), then these can be consistent design point with an effective
combined (Castillo et al 2005) to find the partial factor for other components of the
√6 distribution of c∕ck . design load cd ∕ck ≈ 1.2 and design wind
θ(p) = – [γ + ln(– ln(1 – p))](16)
π Botha et al (2018a) considered four speed annual exceedance probability
different sets of ci ∕cki distributions, result- pt:vd ≈ 2 × 10 –3 (500-year return period).
The vw ∕vw50 would still be Gumbel distrib- ing in four probabilistic models. The first Given that Botha et al (2018a) did not
uted, with mean: model (New-SANS) was the result of a advocate the direct use of the New-SANS
detailed study into wind load uncertainties model, the design point indicated by the
vw μν w by Botha (2016), who found that uncertain- updated models is accepted as repre-
E = (17)
vw50 μνw + θ(0.02)σ vw ties in c∕ck could be significantly higher sentative of the overall reliability currently
1 than typically assumed. offered by SANS 10160-3:2019.
=
1 + 2.59δvw Botha et al (2018a) was concerned that,
given the significant increase in variability Optimal estimation procedure
and standard deviation of c∕ck, using only the New-SANS model If only the recordings from a particular
in a reliability assessment may be too site are used to estimate vd , then more
vw σ νw drastic. Instead, a Bayesian philosophy was representative values are obtained on aver-
STD = (18)
vw50 μνw + θ(0.02)σ vw employed, where the New-SANS model age (low bias), although this introduces
δv w was taken as a prior and used to update substantial statistical uncertainty (high
=
1 + 2.59δvw three other existing c∕ck models by Holicky΄ variance) (Hong et al 2016). A lower vari-
(2009), Milford (1985) and Gulvanessian ance alternative is to regionally aggregate
where δvw = σ νw∕μνw is the coefficient of and Holicky΄ (2005). The ci ∕cki components the data (Hosking & Wallis 2005), although
variation of vw. of each of Botha et al (2018a)’s four models this risks higher bias.
Botha et al (2018a) used a δv 2 = 0.31 to (New-SANS and the three updated models) Bakker (2021) considered this problem
represent the variation of v 2. This annual were assumed to be log-normally distrib- in terms of the bias variance trade-off
coefficient of variation is equivalent to uted and combined to give four log-normal (Friedman et al 2001). It was found that,
δv1.6 = 0.25. The distribution of v1.6 ∕v50
1.6 c∕ck distributions, with means E[c∕ck] and by employing an optimal combination of
can therefore be reasonably represented standard deviations STD[c∕ck] given in site and regional statistics, a design value
by a Gumbel distribution with mean Table 1. which better corresponds to pt:vd could be
E[vw ∕vw50] = 0.6 and standard deviation found. This optimal estimate v̂d is found
STD[vw ∕vw50] = 0.15 defined by Equations 17 FORM results using sample estimates of mean μ̂v1.6 and
and 18. Using the limit state function (Equation 13) standard deviation σ̂ v1.6 from the n precon-
and defined v1.6 ∕v1.6
50 and c∕ck distributions, a ditioned annual maximum wind speeds
Distribution of c FORM analysis was performed. The results measured at a particular site, located in a
Botha et al (2018a) specified each c∕ck for each of Botha et al (2018a)’s probabilis- region with an average coefficient of vari-
component ci ∕cki as having a normal dis- tic models are given in Table 1. ation δ v1.6 , as:
tribution, with mean E[ci ∕cki] and standard For all models these results imply that
1 ∕1.6
deviation STD[ci ∕cki]. Botha et al (2018a) the wind load dominates the uncertainty 1 + θt δ v1.6
v̂ d = (μ̂v1.6 + θk:opt σ̂v1.6) (19)
was mainly concerned with estimating of the overall formulation, with the annual 1 + θk:opt δ v1.6
E[ci ∕cki] and STD[ci ∕cki] rather than the exceedance probability of the design wind
specific distribution of c∕ck. load βq between 78% and 96% of the annual where

30 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 δ2v1.6(1 + θt δv1.6)(2λ – 4δv1.6) –
4(δv1.6 – δv1.6)2nθt
θk:opt =
δ2v1.6(1 + θt δv1.6)(2δv1.6 λ – κ + 1) –
4(δv1.6 – δv1.6)2n
In this formulation θt = θ(pt:vd) is the
standard score (i.e. the number of
standard deviations away from the
mean) of the design value, and the λ and
κ are the skewness and kurtosis of the
Gumbel distribution, equal to 1.14 and 5.4
respectively.
The δ v1.6 can be estimated by weight-
ing the estimated coefficient of variation
of each of the N series δ̂ v1.6i by its sample
size ni as suggested by Hosking and
Wallis (2005):
δ̂ v1.6 =
∑N
i=1ni δ̂ v1.6 i
(21)
∑
N
i=1ni
and if N is large enough (≈> 20), then
δ v1.6 ≈ δ̂ v1.6 . Bakker (2021) recommended
that thunderstorm and synoptic dominated
areas be considered as separate regions
with their own δv = 1.6. The 83 thunder-
storm series yield δ v1.6 = 0.25 and the 87
synoptic series yield δ v1.6 = 0.18.
An estimate of the design wind speed
v̂d can thus be found for a pt:vd . This can be
expressed as v̂d = f(pt:vd , Θ) where f denotes
that v̂d is a function of pt:vd and some other
parameters Θ dependent on the available
data. In sites with a single mechanism
Θ = {μ̂v1.6 , σ̂v1.6 , n, δ v1.6} and in sites with two
mechanisms the parameters from both
need to be considered Θ = {Θ1, Θ2}. The v̂d
at sites with more than one mechanism can
be found by accounting for multiple mech-
anisms, as recommended by Gomes and
Vickery (1978), described in Appendix A.
The procedure outlined in this section
provides a solution for the estimation of
vd at weather stations throughout South
Africa, but it is still unclear how to extend
this information for the entire country and
incorporate it into the standard.
UPDATING DESIGN WIND
LOADS IN SANS
The current wind loads in SANS 10160-
3:2019 are the culmination of a significant
research effort that includes detailed
surface roughness correction (Kruger et al
2011), climatic classification (Kruger et al
2010), statistical modelling (Kruger et al
2013a), mapping considerations (Kruger et
al 2013b; 2017), and expert engineering and
climatological knowledge.
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
(20)
30
Number of stations
20
10
Mean
0
–0.6 –0.3
Figure 5 Difference in estimated design load and the load currently specified in SANS 10160-3:2019
The estimation procedure outlined
above with some interpolation procedure
(Ye et al 2015) could be used to replace
the current map, although much of the
information contained in it would be lost.
Alternatively, the current map could be
used as a basis from which changes are
recommended where new data (Bakker &
Viljoen 2019) and statistical techniques
(Bakker et al 2021; Bakker 2021) provide a
strong enough indication that the current
values are inappropriate. This approach
will incorporate the new information and
provide a solution to the extension of the
design values throughout the country,
while maximising the amount of expert
knowledge retained.
Comparison of estimated
design loads to SANS
Using the pt:vd = 2 × 10 –3 design fractile
and the statistics from the available data
Θ, vd can be estimated at each station
using the optimal estimation proce-
dure described in the previous section
v̂d = f(2 × 10 –3, Θ). An estimated design
load q̂d for each station can then be
obtained using Equations 1 and 10:
1 ⎫ cd ⎫⎫ v̂2d ⎫ 2
ρ⎪ ⎪⎪ ⎪ck v 50(22)
2 ⎭ ck ⎭⎭ v 250 ⎭
q̂d =

1 q̂d may not always result in the most appro-
= ρ1.2ck v̂2d
2 priate design loads.
This can be compared to the design
wind load currently specified by
SANS 10160-3:2019:
0 0.3 0.6
∆ sans
1
qd:sans = ρ1.6ck v 250(23)
2
where v50 is found using the characteristic
wind map in SANS 10160-3:2019 (Figure 1).
The difference between the predicted q̂d
and the current qd:sans design load can be
assessed as:
q̂d – qd:sans
∆ sans = (24)
qd:sans
1.2v̂2d – 1.6v 250
=
1.6v 250
The distribution of ∆ sans values from all
the stations is summarised in Figure 5.
This shows that SANS 10160-3:2019 is
conservative overall, with the average
∆ sans = −10%, which could be attributed
to the conservative measures taken by
Kruger (2011) or the use of wind speed
intervals in the map (Botha et al 2018b).
There are q̂d values that differ significantly
from qd:sans, suggesting that some could be
reconsidered.
Identification of stations with
unacceptable design wind speeds
The accuracy of v̂d is dependent on the
quantity and quality of the v data used.
Therefore, updating qd:sans to closely match
Large differences between qd:sans and
q̂d present evidence that qd:sans should
be changed, but stations with more data
present stronger evidence because there
31
is higher confidence in these estimates.
To determine at which stations there is 1.0
r
enough evidence to reconsider qd:sans a
statistical test was developed. This is based
on the null hypothesis H0 that the current 0.8

design value is correct:

0.6

P(v̂d:sim < vd)

H0: q̂d – qd:sans = 0 (25)

1.6v̂ d2
If the pt:vd = 2 × 10 –3 design fractile is
0.4
accepted, then this H0 is equivalent to:

1.2v̂ 250
H0: 1.2v̂2d – 1.6v 250 = 0 (26)
0.2

A statistical test with this null hypothesis

means sufficient evidence must be pre-
0
sented before modification of the current 2 000 2 250 2 500 2 750 3 000
standard is considered. 1.2v 2d
This test is founded on the assumption Decrease No change Increase
that v1.6 is Gumbel-distributed and
that μˆ v 1.6 and σˆ v 1.6 are fairly unbiased Figure 6 Empirical cumulative distribution of simulated design values, m = 10 000
estimators of the mean and standard
deviation (Bakker et al 2021; Bakker
2021). To test H0, a set of simulated v̂d Increase Decrease
P-value = 0.001 P-value = 0.2
values is obtained through Monte Carlo P-value = 0.01 P-value = 0.1
simulation: P-value = 0.05 P-value = 0.05
P-value = 0.1 P-value = 0.01
1. For a series of n representative wind P-value = 0.2 P-value = 0.001
speed v observations available at a
station, it is assumed that v1.6 follows
a Gumbel distribution with mean μ̂v1.6
and standard deviation σ̂v1.6 equal to the
sample estimates at the station.
2. Then n simulated wind speeds vsim are
drawn from this Gumbel distribution.
3. The mean μ̂v1.6:sim and standard devia-
tion σ̂v1.6:sim of vsim are estimated.
4. The Θsim = {μ̂v1.6:sim, σ̂v1.6:sim, n, δ v1.6} is
thus available. For stations in a mixed
climate, steps 1 to 3 are performed
for each separate mechanism and
Θsim = {Θsim:1, Θsim:2}.
5. A simulated v̂d is then v̂d:sim = f(pt:vd,
Θsim).
6. Steps 2 to 5 are repeated m times to
obtain a set of m simulated v̂d:sim values. Figure 7 E vidence for increasing or decreasing the current wind speeds in SANS 10160-3:2019;
The proportion r of simulated design val- triangles are plotted at station locations
ues which would result in a load less than
qd:sans is then calculated as: considered. The r value also indicates Recommended changes
whether an increase r < α*∕ 2 or decrease to SANS 10160-3
1 m
1: if (1.2v 2 < 1.6v 250 r > 1 – α*∕ 2 in qd:sans should be considered.
���

���

Bakker and Viljoen (2019) described the

r=
m
∑ 0: otherwise
d:sim (27)
To demonstrate application of the quality control of the data. This involved
i=1
test, an example showing how r, v̂d, and visual inspection of each gust to remove
If m is large enough then a P-value can be v50 relate to the empirical distribution of erroneous measurements, surface rough-
found: v̂d:sim at the Worcester station is shown ness correction, and accounting for
in Figure 6, where m = 10 000 was used. the dynamic response of data recorded
P-value = 1 – |2r – 1| (28) An r value of 0.995 was obtained, which using a Dines anemometer. Bakker
means that the current design value can be and Viljoen (2019) did not account for
A significance level α* can be selected and rejected at a high significance level and that topographical features or obstructions
if P-value < α*, then H0 can be rejected, a decrease in the prescribed wind speed around instruments, so some stations
and a change of qd:sans at that station could be considered. with poor exposure cannot be considered

32 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 be a somewhat subjective process. As a
starting point, α* = 0.1 is used to inform
a first round of updates. The current v50
values are thus increased or decreased to
an updated value v50:updated as:

v50 + 4: if r < 0.05

�����
v50:updated = v50 – 4: if r > 0.95 (29)
‑v50: otherwise.

It was found that 42 stations had a

Figure 8 Updated characteristic map (v50:updated); municipalities with data are highlighted
representative of typical climatic condi-
tions. Of the 132 stations, 102 have data
with acceptable exposure.
Figure 7 overlays the results of the
hypothesis test developed in the previ-
ous section applied to each of these 102
stations onto the current characteristic
map. On this figure the shape and colour
indicate whether an increase or decrease
would be appropriate (red triangles vs
blue inverted triangles) according to r,
and the size of the symbol indicates the
strength of the evidence against the cur-
rent value (P-value). This figure will be
used to guide recommended changes to
SANS 10160-3:2019.
Figure 7 shows that, while there are
some isolated municipalities where an
increase could be considered, notably the
eThekwini (Durban) metropolitan munici-
pality, there is strong evidence to support
the reduction of qd:sans throughout much
of South Africa. This is especially clear
in the southwest of the country where
synoptic winds dominate the extreme
wind climate, such as the western parts of
the Western Cape and a stretch along the
Eastern Cape coast. The indication of a
decrease in these areas can be attributed
to the lower coefficient of variation in
these areas, which means that γ w = 1.6
was probably overly conservative in
these regions.
For 150 of the 234 municipalities
there are no usable weather stations,
and so no wind speed data is currently
Wind speed
32 m/s
36 m/s
40 m/s
44 m/s
available. These municipalities tend to
be rural, with limited development, and
thus the lack of data is not a major issue as
SANS 10160-3:2019 aims to provide for the
built environment.
To improve the standard, different
qd:sans values which better represent the
data could be recommended. The areas
where change could be considered can be
identified using Figure 7. To streamline the
revision process and simplify implementa-
tion for practising engineers, it would be
desirable if these changes can be made
without significantly modifying the current
format of the standard.
Specifying vd values per reliability class,
as done in ASCE 7-16 (ASCE 2017) and AS/
NZS 1170.2 (AS/NZS 2011), would require
a substantial modification to the format of
SANS 10160-3:2019. Synoptic and thunder-
storm winds, which dominate South Africa,
are more similar to each other than cyclonic
winds (Holmes 2018). Therefore, adjusting
the design value from the reference reliabili-
ty (annual βt = 4.0) (Retief & Dunaiski 2009)
to other classes should not introduce sig-
nificant error and only the vd corresponding
to βt = 4.0 will be considered. This means
that modification to the standard could be
implemented by making adjustments to the
characteristic wind speed map (Figure 1).
New characteristic wind map
There are a number of valid approaches
which could be used to update the char-
acteristic map, and therefore updating will
P-value < 0.1, indicating that at these loca-
tions there is reasonable justification to
change the qd:sans. For the majority of these
stations a decrease in qd:sans could be con-
sidered, with r > 0.95 at 37 stations.
Some municipalities have multiple
stations. In this case, a change was only
implemented if it could be supported at all
the stations in the municipality. This was
the case for Buffalo City (East London),
eThekwini (Durban), and Saldanha Bay.
If stations in a municipality contradicted
each other, then the more conservative
option was taken and only increasing v50
was considered. This was the case for
the City of Cape Town and Msunduzi
(Pietermaritzburg) where the current v50
values were maintained despite one sta-
tion indicating that a decrease could be
appropriate.
The v50:updated values are shown in
Figure 8 where municipalities with data
are highlighted. Because the selection of
α* = 0.1 is somewhat arbitrary, this updated
map is only meant to serve as a foundation
for further changes.
The Koingnaas station located in the
Kamiesberg municipality indicates that
a decrease could be considered at a 10%
significance level, but here v50 = 32 m/s
and could not be decreased without creat-
ing another v50 category. Since this is an
isolated case, another category was not cre-
ated, and the current value was maintained
at v50:updated = 32 m/s.
A noticeable difference between the
updated map versus the current map is
the reduction in municipalities with the
highest v50 = 44 m/s wind speed category.
There is only one station (Elliot) for which
v50 = 44 m/s cannot be rejected at a 10%
significance level. The data from this
station still indicates that a decrease
would not be unreasonable with r = 0.92.
Therefore, without reasonable justification
to maintain the v50 = 44 m/s category, it is
recommended that it be eliminated.
Historical data, recorded using a
Dines anemometer, is available for several

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 33
stations. This data is of lower quality
due to uncertainty in the accuracy of the
correction applied by Bakker and Viljoen
(2019). Inclusion of the Dines data results
in less variation of v 2d:sim values which leads
to lower P-values. Given these uncertainties
in the quality of this data, it alone was
not considered sufficient evidence to
modify qd:sans. It was checked whether the
more recent data, for which there is more
certainty in the exposure and instrument
response, could support a change
independent of the Dines data. If it could
not, as was the case at the Cape Town
station, then changes were not made. Wind speed
Kruger et al (2017) assigned the current 32 m/s
36 m/s
v50 values to municipalities without data, 40 m/s
using interpolation and his knowledge of
climatology. Therefore, it is argued that the
v50 values at these municipalities can be
changed to create a smoother map where
necessary. There are also some isolated
municipalities where the v50:updated value Figure 9 New characteristic wind speed map, recommended for future update to SANS 10160-3
is surrounded by stations which generally
support a different v50:updated value. These
irregularities were removed by adopting the
more common v50:updated from surround-
ing stations.
The smoothing and elimination of
v50 = 44 m/s in the updated map yielded
a final recommended characteristic wind
speed map, shown in Figure 9. The new
map results in changes to 63 of the 234
municipalities throughout the coun-
try (Figure 10), of which 54 involve a
decrease in v50.
There are eight metropolitan munici-
palities in South Africa. These contain
a large proportion of the population and
built environment of the country, which
means changes to SANS 10160-3:2019 in
these areas are of higher consequence.
Decrease
The new map recommends changes for Increase
the eThekwini (Durban) and Buffalo City No change
(East London) metropolitan municipali-
ties. In both eThekwini and Buffalo City
a P-value < 0.1 was observed at two sta- Figure 10 Differences between current and new characteristic wind speed maps
tions, which is considered to be sufficient
evidence that the recommended changes by directly considering the design values standard was investigated. This was found
for metropolitan municipalities could rather than characteristic values to utilise to be inappropriate, as it made the assump-
be adopted. more information and better attain the tion that all wind loading uncertainties
target reliability. could be attributed to wind speeds, which
This study does not aim to modify the does not match the assumptions used
CONCLUSIONS reliability level of the standard. Thus, to to derive the current loads in the South
The South African Weather Service has avoid recalibration, the annual design wind African standard.
collected a large quantity of wind speed speed exceedance probability currently Instead, the wind load uncertainty
data which has been organised into a provided by the standard could be used to models used to calibrate the current stand-
dataset appropriate for assessment of the estimate design wind speeds. ard were applied to calculate the exceed-
South African wind loading standard. This To find this exceedance probability, ance probability of a design wind speed
data can be incorporated into the standard an approach used for the American ASCE and an effective partial factor for other

34 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
components of the design load. Together
these were used to estimate the design
wind load using the available data and an
optimal procedure that combines wind
speed data collected at a specific site with
regional averages.
Deviation of the estimated design
values from the current design values
provides evidence in favour of changing
the current design values, although the
strength of this evidence is dependent on
the quantity and quality of the data used.
To determine where there was sufficient
evidence that the current design values
were inappropriate, a statistical hypothesis
test was developed.
This test indicated that changes to the
current design values could be considered
at a number of sites throughout the coun-
try. These changes were implemented in
a new characteristic map of basic wind
speed (Figure 9) that should lead to a better
reflection of the South African climate in
the standard.
REFERENCES
Ang, A H-S & Tang, W H 1984. Probability concepts
in engineering planning and design. Vol 2. Decision,
risk, and reliability. New York: Wiley, p 608.
ASCE (American Society of Civil Engineers) 2017.
ASCE 7-16. Minimum Design Loads and Associated
Criteria for Buildings and Other Structures. Reston,
VA: ASCE.
AS/NZS (Australian Standard/New Zealand
Standard) 2011. AS/NZS 1170.2 (2011). Structural
Design Actions. Part 2. Wind Actions. Joint
Technical Committee BD-006, Australia/New
Zealand.
Bakker, F P & Viljoen, C 2019. An analysis of South
African wind gust data in the context of the built
environment. Proceedings, 7th International
Conference on Structural Engineering, Mechanics
and Computation (SEMC 2019), Boca Raton,
FL: CRC Press, pp 2352–2358.
Bakker, F P, De Koker, N & Viljoen, C2021.
Preconditioning wind speeds for standardised
structural design. Engineering Structures, 238,
111856.
Bakker, F P2021. Characterisation of the South
African extreme wind environment relevant
to standardisation. PhD Thesis. Stellenbosch
University.
Baravalle, M & Köhler, J 2018. On the probabilistic
representation of the wind climate for calibration
of structural design standards. Structural Safety,
70: 115–127.
Botha, J 2016. Probabilistic models of design wind
loads in South Africa. PhD Thesis. Stellenbosch
University.
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
Botha, J, Retief, J & Viljoen, C 2018a. Reliability
assessment of the South African wind load design
formulation. Journal of the South African Institution
of Civil Engineering, 60(3): 30–40.
Botha, J, Retief, J V & Viljoen, C 2018b. Uncertainties
in the South African wind load design formulation.
Journal of the South African Institution of Civil
Engineering, 60(3): 16–29.
Castillo, E, Hadi, A S, Balakrishnan, N & Sarabia, J M
2005. Extreme Value and Related Models with
Applications in Engineering and Science. Hoboken,
NJ: Wiley.
Cook, N 1985. The designers’ guide to wind loading of
building structures. Part 1. Watford, UK: Building
Research Establishment.
Cook, R, Griffis, L, Vickery, P & Stafford, E 2011.
ASCE 7-10 wind loads. Proceedings, Structures
Congress 2011, 14–16 April 2011, Las Vegas, NV,
pp 1440–1453.
Davenport, A 1982. The interaction of wind
and structures. In Plate, E J (Ed.), Engineering
Meteorology. Amsterdam: Elsevier, pp 527–572.
Ellingwood, B R & Tekie, P B 1999. Wind load statistics
for probability-based structural design. Journal of
Structural Engineering, 125(4): 453–463.
EN (European Standard) 2005. EN 1991-1-4 2005.
Eurocode. Actions on Structures. Parts 1-4: General
Actions – Wind Actions. Brussels: European
Committee for Standardisation (CEN).
Friedman, J, Hastie, T & Tibshirani, R 2001. The
Elements of Statistical Learning. Vol 1. New York:
Springer Series in Statistics.
Goliger, A M, Retief, J V & Kruger, A C 2017. Review
of climatic input data for wind load design
in accordance with SANS 10160‑3. Journal
of the South African Institution of Civil
Engineering, 59(4): 2‒11.
Gomes, L & Vickery, B 1978. Extreme wind speeds in
mixed wind climates. Journal of Wind Engineering
and Industrial Aerodynamics, 2(4): 331–344.
Gulvanessian, H & Holicky΄, M 2005. Eurocodes:
Using reliability analysis to combine action effects.
Proceedings of the Institution of Civil Engineers –
Structures and Buildings, 158(4): 243–252.
Holicky΄, M 2009. Reliability Analysis for Structural
Design. Stellenbosch: African Sun Media.
Holmes, J D 2018. Wind Loading of Structures. Boca
Raton, FL: CRC Press.
Hong, H, Mara, T, Morris, R, Li, S & Ye, W 2014.
Basis for recommending an update of wind velocity
pressures in Canadian design codes. Canadian
Journal of Civil Engineering, 41(3): 206–221.
Hong, H P, Ye, W & Li, S H 2016. Sample size effect
on the reliability and calibration of design wind
load. Structure and Infrastructure Engineering,
12(6): 752–764.
Hosking, J R M & Wallis, J R 2005. Regional Frequency
Analysis: An Approach Based on L-Moments.
Cambridge, UK: Cambridge University Press.
JCSS (Joint Committee on Structural Safety) 2001.
JCSS Model Code (2001 February). Probabilistic
Model Code. JCSS Standard.
Kruger, A C, Goliger, A M, Retief, J V & Sekele, S
2010. Strong wind climatic zones in South Africa.
Wind and Structures, 31(1): 37‒55. DOI:10.12989/
was.2010.13.1.037.
Kruger, A C 2011. Wind climatology of South Africa
relevant to the design of the built environment. PhD
Thesis. University of Stellenbosch.
Kruger, A, Retief, J & Goliger, A M 2011. An updated
description of the strong-wind climate of South
Africa. Proceedings, 13th International Conference
on Wind Engineering, 10–15 July 2011, Amsterdam,
Netherlands.
Kruger, A, Retief, J & Goliger, A M 2013a. Strong winds
in South Africa. Part 1. Application of estimation
methods. Journal of the South African Institution of
Civil Engineering, 55(2): 29–45.
Kruger, A, Retief, J & Goliger, A M 2013b. Strong
winds in South Africa. Part 2. Mapping of updated
statistics. Journal of the South African Institution of
Civil Engineering, 55(2): 46–58.
Kruger, A C, Retief, J & Goliger, A 2017. Development
of an updated fundamental basic wind speed map
for SANS 10160-3. Journal of the South African
Institution of Civil Engineering, 59(4): 12–25.
Milford, R 1985. Extreme value analysis of South
African gust speed data. CSIR Unpublished Internal
Report, 85/4. Pretoria: CSIR.
Milford, R V 1987. Annual maximum wind speeds
for South Africa. Civil Engineering/Siviele
Ingenieurswese, 1987: 15–19.
Palutikof, J, Brabson, B, Lister, D & Adcock, S 1999.
A review of methods to calculate extreme wind
speeds. Meteorological Applications, 6(2): 119–132.
Retief, J V & Dunaiski, P 2009. Background to SANS
10160: Basis of Structural Design and Actions for
Buildings and Industrial Structures. Stellenbosch:
African Sun Media.
SANS (South African National Standard) 2019a. SANS
10160-1. Basis of Structural Design and Actions for
Buildings and Industrial Structures. Part 1. Basis of
Structural Design. Pretoria: SABS Standards Division.
SANS (South African National Standard) 2019b. SANS
10160-3. Basis of Structural Design and Actions for
Buildings and Industrial Structures. Part 3. Wind
Actions. Pretoria: SABS Standards Division.
Ye, W, Hong, H & Wang, J 2015. Comparison of spatial
interpolation methods for extreme wind speeds over
Canada. Journal of Computing in Civil Engineering,
29(6): 04014095.
APPENDIX A
A series of annual maximum wind speeds
may not be identically distributed, because
the extreme wind environment could
be a result of m different extreme wind
mechanisms.
35
 The v fractile vjoint that results from the
joint distribution of m mechanisms for a
given exceedance probability p is unknown.
This can be found if the extreme wind
speed record can be segregated by mecha-
nism, i.e. an extreme from each mechanism
extracted per block and the joint distribu-
tion considered, as recommended by
Gomes and Vickery (1978).
36
The vjoint is equal to the fractile of each
marginal distribution vi:
vjoint = v1(p1) = v 2(p2) = … vm(pm)(A1)
Provided each mechanism is independent,
p can be related to the exceedance prob-
ability of each independent mechanism
pi as:
Volume 64 Number 1 March 2022
(1 – p) = (1 – p1)(1 – p2) … (1 – pm)(A2)
These m unknown pi values and the
unknown vjoint constitute (m + 1) unknowns.
Equations A2 and A3 yield a system of
(m + 1) equations. Therefore, provided each
vi can be related to pi, the (m + 1) equations
can be used to solve the (m + 1) unknowns.
This solution will provide the vjoint for a
given p, as is typically required.
Journal of the South African Institution of Civil Engineering
 TECHNICAL PAPER
Assessment of local
sewage sludge ash as a Journal of the South African
Institution of Civil Engineering

supplementary cementitious
ISSN 1021-2019
Vol 64 No 1, March 2022, Pages 37–47, Paper 1315

material – effects of RUTH KATLEGO JUALA works as an Engineer

in Training at Sasol’s Research and

incineration temperature
Technology Department. She holds BSc in
Metallurgical and Materials Engineering and
MSc in Chemical Engineering (obtained

and cooling rate of the ash

with distinction) degrees from the
University of the Witwatersrand. This paper
is based on the research study that she
undertook for her MSc degree. Her research and intellectual interests are
in reduction of greenhouse gas emissions, alloy waste and the production
R Juala, Y Ballim, J Mulopo of sustainable synthetic products through the integration of solid waste
and biomass materials.

Contact details:
School of Chemical and Metallurgical Engineering
This paper reports on the possible use of sewage sludge ash as a pozzolanic supplementary
University of the Witwatersrand
cementitious material to Portland cement. Samples of sewage sludge were incinerated at Private Bag 3, Wits 2050, South Africa
700ºC, 800ºC and 900ºC and these were then cooled in the furnace (FISSA), in air (AISSA) or by T: +27 73 216 9312
quenching in water. The resulting ashes were ground to suitable fineness and used to prepare E: katlegojuala@gmail.com

cement pastes and mortars in which the binder consisted of 30% ash and 70% Portland cement.
PROF YUNUS BALLIM is an Emeritus
The paste samples were used for microscopic and chemical assessment of the evolution
Professor in the School of Civil and
of hydration products, while the mortars were used to assess the effects of the ashes on Environmental Engineering at the University
workability and compressive strength of laboratory-prepared samples using a water/binder of the Witwatersrand. His research is mainly
ratio of 0.5. Fly ash was used as a reference pozzolanic material to assess the performance in cement and concrete materials science
and engineering, and he has held a National
of sewage sludge ashes. Analysis of the sewage sludge ashes showed the presence of
Research Foundation rating as a researcher
cementitious compounds and hydration products that suggest that this material can be used as since 1994. He is an Honorary Fellow of
a partial replacement of Portland cement. However, sewage sludge ash reduces the workability SAICE and a Senior Member of RILEM (International Union of Laboratories
of the mortar. Compressive strength results indicate that the highest strength is obtained when and Experts in Construction Materials, Systems and Structures). He has
served on a number of international technical committees related to
the sewage sludge is incinerated at 900ºC and then quenched in water. cement and concrete materials research.

Contact details:
School of Civil and Environmental Engineering
INTRODUCTION historically been centred around the University of the Witwatersrand
Waste disposal is a recognised global presence of harmful pollutants such as Private Bag 3, Wits 2050, South Africa
T: +27 11 717 7103
health problem and, with the human popu- pathogens (e.g. viruses, parasites and
E: yunus.ballim@wits.ac.za
lation growing, the demands for safe waste bacteria); organic contaminants (e.g.
disposal are increasing. Sewage sludge, dioxins, surfactants and pharmaceuticals) PROF JEAN MULOPO is an Associate
which is the residue from a wastewater and inorganic contaminants (e.g. metal Professor at the University of the
treatment plant, is one of the largest and trace elements) (National Research Witwatersrand’s School of Chemical and
Metallurgical Engineering, where he
particularly problematic contributors to the Council 2002; Lewis & Gattie 2002).
conducts research in the areas of
burden of waste disposal worldwide. The There is clearly a need for much further waste-to-energy, waste-to-resources, solid
sludge is of a semi-solid nature produced as assessment and analysis of alternative use waste and wastewater treatment,
a by-product from treatment of municipal possibilities to develop the evidence base bioengineering, and sustainable process
development (http://orcid.org/0000-0001-9786-6799). His research
sewage and industrial wastewater. The that will improve sewage sludge disposal
focuses on the development of low-cost waste and wastewater treatment
global estimate is that between 70 and 105 practices around the world. technologies that enhance the recovery and reuse of valuable products
million tons of sewage sludge is produced In its turn, the concrete industry has such as water, nutrients, and chemicals.
annually (Grobelak et al 2019). Waste long advocated for increased use of sup- Contact details:
recycling in the construction industry is plementary cementitious materials (SCM) School of Chemical and Metallurgical Engineering
one of the best ways to provide new, safe such as ground-granulated blast furnace University of the Witwatersrand
Private Bag 3, Wits 2050, South Africa
and renewable resources that contribute to slag and fly ash (FA) to replace a part of the T: +27 11 717 7507
economic development and thus protect Portland cement used in concrete and so E: jean.mulopo@wits.ac.za
the environment. reduce the general carbon dioxide footprint
Fleishman et al (2014) argue that of concrete construction (Scrivener et al
South Africa is one of the countries facing 2017). Incinerated sewage sludge ash (ISSA)
difficulties in the disposal of wastewater reportedly contains oxides such as Al2O3,
sludge. Environmental and public health CaO and SiO2 which are the primary Keywords: sewage sludge ash, pyro-processing, pozzolanicity,
concerns of sewage sludge disposal have compounds found in Portland cement (PC) mortar strength, workability

Juala R, Ballim Y, Mulopo J. Assessment of local sewage sludge ash as a supplementary cementitious material – effects of incineration temperature
and cooling rate of the ash. J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #1315, 11 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64no1a4 37
(Tenza-Abril et al 2015), which makes ISSA
a potential SCM or pozzolanic material
that can be used in cement-based materi-
als (Tay 1987; Gomes et al 2019; Payá
et al 2019).
Previous studies (Monzó et al 1997;
Tantawy et al 2012; Dyer et al 2011; Pavlik
et al 2016) have shown evidence that ISSA
exhibits a certain level of pozzolanic activ-
ity, as is the case with fly ash and silica
fume. Pozzolans are materials that contain
aluminous and siliceous material, which
when finely divided react with calcium
hydroxide Ca(OH)2 to form cementing
compounds found in hardened cement
paste (Massazza 1998). The use of these
materials in concrete brings durability
benefits while reducing the CO2 footprint
of the binder and reducing the amounts of
waste material sent to landfill (Scrivener
et al 2017).
The decomposition of sewage sludge
with increasing temperature occurs at dif-
ferent stages. The four principal stages are
described below (Tantawy et al 2012):
QQ The loss of moisture and absorbed
water at 50ºC ‒ 130ºC
QQ The emission of volatile organic matter
at 200ºC ‒ 320ºC
QQ The combustion of organic content
due to the varying boiling points of
hydrocarbons in the sewage sludge at
330ºC ‒ 390ºC
QQ The thermal decomposition of fixed
carbon captured by inorganic matter,
loss of structural water of the clay
minerals in the sewage sludge and the
elimination of carbonaceous matter in
the sewage sludge at 400ºC ‒ 670ºC.
To ensure that all organic matter is fully
decomposed before the material is used as
an SCM in concrete, sewage sludge should
therefore be incinerated at a minimum
temperature of 700ºC.
Figure 1 shows a calcium, silica and
alumina oxide ternary diagram of cementi-
tious materials commonly used in concrete.
The chemical analysis of materials in
this study was used to locate ISSA on the
diagram. It can be seen that ISSA overlaps
strongly with FA in terms of the CaO, SiO2
and Al2O3 content.
Previous studies have shown that the
partial replacement of PC by ISSA affects
both the strength and workability of binder
pastes (Pan et al 2003; Monzó et al 2003).
Other studies have shown that using ISSA
as an SCM in mortar extends the setting
time and reduces the workability of the
mortar (Vouk et al 2016; Naamane et al
38
SiO2
Silica
fume
Fly ash
ISSA
Slag
Portland cement
Fine limestone
CaO
Figure 1 T ernary diagram of silica, alumina and calcium oxides showing commonly used
cementitious materials and the relative position of ISSA used in this study (adapted from
Rakhimov et al 2015)
2016). The increased setting times hav been
attributed to the high amounts of P2O5 and
SO3 found in the ISSA (Vouk et al 2016;
Naamane et al 2016).
Vouk et al (2016) investigated the effect
of ISSA on the workability of cement
mortar. The results showed that 10%
replacement of cement by ISSA caused a
19% reduction of the mortar workability.
Similar results were obtained by Monzó et
al (1996) who used ashes with three dif-
ferent fineness levels, at 15% replacement
of PC. This reduction in workability of the
cement mortar was attributed to the irreg-
ular morphology and the increased water
absorption of ISSA. Vimonsatit et al (2015)
point out that the significantly higher water
demand of ISSA can be resolved by adding
coal fly ash.
Naamane et al (2016) investigated
the effect of sewage sludge incineration
temperatures on the compressive strength
of mortar produced by partially replacing
cement with ISSA. They found that an
incineration temperature of 800°C for a
period of 2.5 hours yielded the highest
compressive strength of the PC/ISSA mor-
tars after 90 days.
Vouk et al (2017) found that, at 72 days,
mortars containing 90% PC and 10% ISSA
had a compressive strength that was about
7% lower than that of companion mortars
made with 100% PC. In a study conducted
by Cyr et al (2007), the effect of the cur-
ing time on the compressive strength of
Volume 64 Number 1 March 2022
Natural
pozzolans
Metakaolin
AI O3
mortars containing ISSA at replacement
levels of 25% and 50% was considered. It
was found that the rate of compressive
strength gain was lower at early ages,
but accelerated after 28 days. Also, the
compressive strength of mortar containing
25% ISSA was higher than that of mortar
containing 50% ISSA.
The study reported in this paper was
aimed at evaluating the feasibility of ISSA
from a wastewater treatment plant in the
Gauteng area to be used as an SCM in
cement-based materials for construction.
In particular, the study was focused on
the effects of the post-incineration cooling
rate of ISSA on the workability and com-
pressive strength of mortars, when used at
30% replacement of PC. Samples of sewage
sludge were incinerated at 700 oC, 800 oC
and 900 oC and then cooled in the furnace
or in air or by being quenched with water.
The sewage sludge and the resulting ISSA
materials were characterised for their
physical and chemical characteristics.
Paste and mortar samples made with ISSA
and PC were assessed for their micro-
structural development during early
hydration, as well as for workability in the
plastic state and for compressive strength
development up to 28 days. Pastes and
mortars prepared with plain PC and with
a 70/30 blend of PC and fly ash (FA) were
used as reference materials for compari-
son of the overall performance of ISSA as
a possible SCM.
Journal of the South African Institution of Civil Engineering
Table 1 C
 hemical composition (in %) of PC, FA
(Ballim & Graham 2009) and DSS 1 000
Solid lines: Cooled in furnace
900
Oxide PC FA DSS Dashed lines: Cooled in air
800
CaO 64.00 4.20 3.88
700

Temperature (°C)
SiO2 22.90 55.10 21.42
600
Al2O3 4.60 33.30 5.89 500

Fe2O3 2.58 3.15 4.79 400

300
MgO 1.60 1.20 1.12
200
Na2O 0.12 0.00 0.22
100
P2O5 0.06 0.39 3.95 0
0 2 4 6 8 10 12 14 16
K 2O 0.30 0.69 0.51 Time (hr)
TiO2 0.46 1.67 0.44 Maximum heating temperature: 700°C 800°C 900°C

Mn2O3 0.12 0.02 0.13

Figure 2 M
 easured heating and cooling curves for the samples cooled inside the furnace and in
SO3 2.05 0.12 0.00 laboratory air

LOI 2.41 0.45 56.64

Materials and Test Methods
Cementitious materials
In this study PC classified as CEM I, 52,5 N
and FA from a local supplier were used to
prepare paste and mortar samples. Sewage
sludge with 95% solids and 5% water was
collected from a wastewater treatment
plant in Gauteng, after it had been dried
in the sun for some time. The collected
sewage sludge was further dried in an oven
at 100ºC for 24 hours to produce dried
sewage sludge (DSS). X-ray fluorescence
(XRF) analysis was then conducted using a
PANalytical Axios instrument on the DSS
to determine its chemical composition,
and the results are shown in Table 1. The
loss on ignition (LOI) was determined
by roasting the sample at 1 050oC until
constant weight.
As a basis for comparison, the analyses
of samples of PC and FA, similar to that
used in this study and derived from the
same sources, and determined by Ballim
and Graham (2009), are also shown in
Table 1. It is interesting to note that the
DSS used in this study contained no meas-
urable SO3 which Vouk et al (2016) and
Naamane et al (2016) noted as contributing
to increased setting times of the ISSA
materials that they studied. The high LOI
of the DSS is common with organic materi-
als and is mainly caused by a loss of carbon
and bound water.
Separate samples of DSS were then
incinerated at 700ºC, 800ºC and 900ºC in a
closed-lid furnace. The furnace was heated
at an average rate of 17ºC/minute until
700ºC and thereafter, for the 800ºC and
900ºC samples, at an average rate of 4ºC/
minute. The furnace temperature was then
held at the intended incineration tempera-
ture for four hours, after which the furnace
heater was turned off. Three samples of
DSS were prepared at each of the incinera-
tion temperatures and these were cooled
as follows:
QQ One sample was left to cool inside the
furnace.
QQ One sample was immediately removed
from the furnace and left to cool in still
laboratory air.
QQ One sample was immediately removed
from the furnace and quenched by
pouring the ash into a bucket of water
at room temperature. This sample
was then placed in a ventilated oven at
100ºC to dry.
Figure 2 shows the measured heating and
cooling rates of the samples left to cool in
the furnace and in air. The temperature
was monitored by placing a type K ther-
mocouple in the sample, connected to an
external temperature recording unit. The
cooling rate of the quenched sample was
not measured as the temperature drop
occurred too rapidly.
Table 2 shows the yield of ash for the
different incineration temperatures, as a
proportion of the mass of DSS incinerated,
as well as the average cooling rates for the
furnace and air samples, when cooled to
50ºC. This table also indicates the iden-
tification codes for each of the nine ashes
produced for this study.
The cooled samples of ISSA were then
ground for 20 minutes in a pulveriser
(0068I Fact East model). Particle size analy-
sis of the powdered materials was carried
out using a laser diffraction-based Malvern
Mastersizer (MS2000) to determine the
particle size distribution of the PC, FA and

Table 2 Characteristics and identification of the nine ISSA samples produced for this study

Incineration temperature (ºC)

700ºC 800ºC 900ºC

Average ash yield (% by mass) 47 44 41

Cooling method Furnace Air Quench Furnace Air Quench Furnace Air Quench

Average cooling rate (ºC/min) 1.7 3.7 – 1.6 4.3 – 1.4 4.7 –

Sample identification 7F 7A 7Q 8F 8A 8Q 9F 9A 9Q

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 39
 (a)
Figure 3 C
 hange of form of DSS (a), after incineration (b) and after pulverising (c)
ISSA. Figure 3 gives a visual sense of the
transformation of the DSS after incinera-
tion and pulverisation.
Atomic absorption
spectroscopy (AAS) analysis
Atomic absorption spectroscopy was used
to analyse the elemental composition
and concentration of DSS and ISSA to
determine the compositional effects of the
incineration, the incineration temperatures
and cooling rates. A microwave digester
was used to convert pulverised DSS and
ground ISSA samples in an acid suspen-
sion, into a solution prior to the analysis.
Samples for analysis were prepared using
1 g of each material suspended in a com-
bination of 3 mL hydrochloric acid (HCl)
and 10 mL nitric acid (HNO3) standard
solution. The samples were then digested for
20 minutes and filtered using filter paper.
Because of the high concentrations of HNO3
and HCl, 2 mL of the digested and filtered
mixture was diluted with water to produce
a standard solution of 25 mL with a suit-
able concentration level to produce reliable
results. The samples were then analysed
using an Agilent 200 Series AAS analyser.
Acetylene was used as fuel gas and air as the
oxidant to detect each element after excita-
tion by hollow cathode lamps.
X-ray fluorescence (XRF) analysis
The DSS and ISSA materials were also
characterised for their chemical composi-
tion using XRF analysis. Samples were ini-
tially ground to suitable maximum particle
sizes and submitted to a local cement labo-
ratory for analysis as part of their routine
testing programme.
Scanning electron microscopy (SEM)
A Jeol JSM 6400 scanning electron micro-
scope, equipped with energy dispersive
40
(b)
spectroscopy (EDS) capacity, was used
for microscopic examination of the ISSA
samples. EDS allows a chemical element
analysis of particular features of the
sample being viewed in the SEM. The
analysis was used to study the surface
morphology of the ground ISSA, as well as
the nature of hydration products after six
hours and seven days of hydration of the
binder pastes. These times were chosen to
give a sense of the early and more mature
stages of hydration. Because SEM analysis
was intended to be largely qualitative,
this part of the study considered only the
ISSA samples that had been incinerated
at 800 oC.
Characterisation tests
A Malvern Mastersizer (MS2000) was
used to determine the particle size distri-
bution of PC, FA and ISSA. The analysis
was conducted on dry powder samples.
Also, the density of a ground ISSA
sample was determined as 2 285 kg/‌m 3
using a standard pycnometer method
(ASTM 2000).
Compressive strength and
workability tests
Mortar samples were prepared to study the
effect of ISSA on workability and compres-
sive strength at a PC replacement level of
30%. To compare the effectiveness of ISSA,
two reference mortars were prepared ‒ one
containing 100% PC and one containing
a 30/70 blend of FA and PC as binders.
A local crushed andesite sand was used
as aggregate for the mortar. A total of 11
mortar mixes were therefore prepared ‒
one with 100% PC as binder, one with 30%
FA as part of the binder, and nine with 30%
ground ISSA as part of the binder, using
all the ISSAs described in Table 2, with
a water/binder ratio of 0.5. The mixture
Volume 64 Number 1 March 2022
(c)
Table 3 M
 ixture proportions (in grams) of
the mortars used for workability
compressive strength testing
70/30 70/30
Binder type PC
PC/FA PC/ISSA
Water 310 310 310
PC 620 434 434
Fly ash – 186 –
ISSA – – 186
Andesite sand 2089 2027 2025
proportions of the mortars are shown in
Table 3. The sand content was adjusted
to account for the increase in volume of
the FA and ISSA in replacing a part of the
PC. The amounts of materials indicated in
Table 3 were sufficient to produce the nine
cube samples.
After mixing, the workability of each
mortar was assessed using a standard flow
table test (ASTM 2007). The mortar flow
is expressed as a percentage increase in
the diameter of the mortar sample over
the original moulded diameter, after 25
drops of the flow table. Thereafter, nine
50 mm cube samples were prepared of
each mortar mixture for compressive
strength testing at 3, 7 and 28 days after
casting (three samples per test age) to
assess the rate and extent of strength
development of the mortars over the
first four weeks after mixing. The cube
samples in their moulds were stored under
plastic sheeting in the laboratory and were
de-moulded the following day. They were
then cured in water at 22 ± 2oC until the
time of testing. Compressive strength
tests were conducted on a Tinius-Olsen
machine with a capacity of 600 kN, at a
loading rate of 15 MPa/minute.
Journal of the South African Institution of Civil Engineering
 9 100

8 90

80
7
70

Cumulative volume (%)

6
Volume density (%)

60
5
50
4
40
3
30
2
20

1 10

0 0
0.01 0.1 1 10 100 1 000 10 000
Size (μm)
Volume density PC Volume density CFA Volume density ISSA
Cumulative volume PC Cumulative volume CFA Cumulative volume ISSA

Figure 4 P
 article size distributions of the PC, FA and ISSA

Results and discussion
Particle size distribution
Figure 4 shows the particles size distribu-
tions of the PC, FA and ground ISSA used
in this study. FA particles were in the range
of 1.7 μm to 180 μm, while ISSA particles
ranged from 1.7 μm to 180 μm, with a
small cluster of particles in the size range
180 μm to 1250 μm. It is thought that this
is a cluster of particles that are harder and
more difficult to grind and that, at this size
range, is unlikely to contribute significantly
to pozzolanic activity of the ISSA. The size
distribution of the PC seems unusually
large, as it ranges from 0.48 μm to 630 μm,
when the maximum particle size for
cement is typically 100 μm (Osbaeck
& Johansen 1989). It is likely that some
agglomeration of particles occurred during
the analysis.
The results in Figure 4 were used to
determine the median particle size, or D50
value, for the three materials as follows:
QQ FA: D50 = 17.37 μm
QQ PC: D50 = 24.60 μm
QQ ISSA: D50 = 28.3 μm
The particle size distribution of the ISSA
is skewed to the coarse side and is gener-
ally coarser than the FA. This may well
influence the observable reactivity of the
ISSA and further studies will be needed to
consider the effects of fineness of grinding
of ISSA. In other studies, on ISSA, the
material was ground to maximum particle
sizes that ranged from 600 μm to 100 μm
(Halliday et al 2012; Garcés et al 2008;
Coutand et al 2006).
Atomic absorption spectroscopy
(AAS) – elemental analysis
Figure 5 shows the AAS results obtained
from an elemental analysis of the DSS and
the nine ISSA samples. The results show
that DSS has a significant amount of the

480

400
Concentration (mg/L)

320

240

160

80

0
DSS 7F 8F 9F 7A 8A 9A 7Q 8Q 9Q
Material
Ca Si Al Fe Mg Na K P

Figure 5 A
 tomic absorption spectroscopy results of DSS and the nine ISSA samples subjected to different pyro-processing regimes

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 41
 25

20
Concentration (mg/L)

15

10

0
DSS 7F 8F 9F 7A 8A 9A 7Q 8Q 9Q
Material
Ti Zn Mn V Cr Co Cu Pb Ni

Figure 6 H
 eavy metal elements detected in DSS and ISSA

chemical elements found in PC (Ca, Si, Al,
Fe, Mg and Na). As expected, incineration
had a significant effect in increasing the
concentration of elements in the ISSA, since
water and volatile materials are removed
from the sample during heating. Similar
results were noted in previous studies
(Halliday et al 2012; Chen & Lin 2009;
Coutand et al 2006), varying in relation to
the composition of the DSS used. Also, the
lowest concentrations were noted for the
furnace-cooled samples, while the air-cooled
and quenched samples showed similar
concentration results. It is possible that this
is the result of morphological changes that
occur during cooling, as this may be influ-
enced by the cooling rate of the ash.
Potassium and phosphorus were also
detected in small amounts in both the DSS
and the ISSA samples. The potassium con-
centration was reduced as a result of incin-
eration, whereas the phosphorus content
showed a slight increase after incineration.
In consideration of the toxicity and
pollution potential of DSS and ISSA, AAS
was also used to detect the presence of the
following heavy metals: lead (Pb), chro-
mium (Cr), copper (Cu), nickel (Ni) zinc
(Zn), cobalt (Co), Titanium (Ti), vanadium
(V) and manganese (Mn). These results are
shown in Figure 6.
While the concentrations of heavy met-
als in the samples are much smaller than
those of the elements shown in Figure 5,
they are large enough to warrant concern.
Notable increases in concentrations of Ti
and Mn were observed after incineration.
The concentrations of Ni, V and Cr were
largely unaffected by incineration.
After incineration, there were sig-
nificant reductions in Pb, Zn, Cu, and Co
concentrations because these metals are
volatilised as the temperature approaches
their boiling points. The presence of heavy
metals in DSS tends to limit its industrial
application, and the possibility of binding
or encapsulating the elements makes its use
in concrete attractive. However, releasing
some of the elemental content into the
atmosphere presents a serious concern and
points to the need for active removal of

70

60

50
% by mass

40

30

20

10

0
PC FA DSS 7F 7A 7Q 8F 8A 8Q 9F 9A 9Q
Material
Calcium oxide Silicon dioxide Aluminium oxide Ferric oxide
Magnesium oxide Phosphorus pentoxide Potassium oxide Sodium oxide

Figure 7 X
 RF results of PC and FA compared to DSS and ISSA materials

42 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
these elements from the flue gasses during
Ca O Fe Al Si P Ca Fe Fe
incineration to reduce the impact of such
metals on the environment.
C

X-ray fluorescence – oxide

compound analysis
The compositions of the ISSA samples used B
in this study were determined using XRF
A
analysis and are compared to the reference
PC and FA as determined by Ballim &
Graham (2009) in Figure 7. The principal
compounds in the ISSA samples were
SiO2 (49.08% to 53.88%), CaO (6.89% to
8.94%), Al2O3 (13.26% to 15.13%) and Fe2O3 Figure 8 M
 icrographs and EDS of ISSA incinerated at 800°C and furnace-cooled (8F – element
(10.19% to 11.38%). The CaO/SiO2 ratio of identification indicated on top left EDS for clarity)
the ISSA materials ranges from 0.12 to 0.16,
which is slightly higher than that of the FA
(0.08), indicating that ISSA may be suitable O Fe Al Si P Ca Fe Fe

for use amended as an SCM to partially

replace PC. The significant presence of
C
Al2O3, Fe2O3 and CaO can be linked to the
use of ferric salts, aluminum and lime as
flocculators in the secondary sludge con- B
ditioning at wastewater treatment plants
(Halliday et al 2012; Tantawy et al 2012),
A
as well as the oxidation of some of the ele-
ments on exposure to air.
In general, incineration had a minor
impact on the component composition
of the DSS. The ISSA samples that were
burned at 800oC and 900oC revealed the Figure 9 M
 icrographs and EDS of ISSA incinerated at 800°C and air-cooled (8A – element
most significant, albeit modest, increases identification indicated on top left EDS for clarity)
in chemical concentrations. However, no
clear relation exists between the compound
composition of ISSA materials and the K O Fe Al Si P K Ca Fe Fe
D
incineration temperature or cooling rates
used. It should also be noted that the 2
A
DSS and all the ISSA materials contain a 1
significant amount of P2O5 (phosphorus
pentoxide), varying from 7.74% to 9.24%. C
This may be attributed to the domestic
detergents either in the waste or those used
at the wastewater treatment plant. Also, B
although not noted on Figure 7, LOI from
the XRF analyses of the ISSA materials was
in the range of 1.16% to 5.90%, increasing
with decreasing incineration temperature
of the original ashes, and this was noted for Figure 10 M
 icrographs and EDS of ISSA incinerated at 800°C and quenched in water (8Q)
all the cooling methods.
well-developed surfaces and a few traces of Figure 10). Some platy or more prismatic
Morphology and platy, rectangular or near-spherical grains. particles with well-defined edges were
composition of ISSA These results are similar to research find- noted in the samples (see particle B in
Figures 8 to 10 show the morphology of ings on ISSA in other parts of the world Figure 8 and particle C in Figure 9).
ISSA incinerated at 800ºC and cooled in (Monzó et al 1996; Tantawy et al 2012; EDS traces of the particles shown on
the furnace (Figure 8), in air (Figure 9) and Naamane et al 2016). Fine particles were the SEM micrographs indicate high peaks
quenched (Figure 10) after incineration. also observed in all ashes incinerated at of silicon (Si) as the main constituent of
These figures also show the EDS trace different temperatures, and these often the particles assessed, with some por-
analyses of the grains indicated. The appear to adhere to the surfaces of larger tions of the ISSA containing aluminum
results show that ISSA has an irregular particles (see particle C in Figure 8, parti- (Al), calcium (Ca), phosphorus (P) and
morphology and size of grains with cle A in Figure 9, and particles A and B in iron (Fe). These results correlate with the

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 43
composition of the ISSA noted with the
AAS and XRF results.
Traces of diatomite (or diatomaceous
earth) were also noted in the ISSA samples
(indicated as particle C in Figure 10). This
was confirmed by both the tubular shape of
the particles as well as the EDS elemental
analysis. Diatomite is a low-cost material
with a variety of properties commonly used
as an adsorbent in wastewater treatment
plants (Zhang et al 2009). A closer view of
a cylindrical diatomite particle is shown
as particle 2 in Figure 10 D, and the EDS
analysis indicates that it is largely com-
posed of silica.
Particle 1 in Figure 10 D illustrates
the irregular morphology of ISSA, more
characteristic of the quenched sample.
Closer views of the morphology of the dif-
ferently cooled ISSA samples are shown in
Figure 11. This figure shows that the rate of
cooling affects the particle morphology of
the ISSA, with faster cooling rates produc-
ing more particles with a darker shade of
grey (denser) and that are generally smaller
and more nodular, rather than platy or
layered. Higher cooling rates produce more
glassy or amorphous forms of silica, which
encourage pozzolanic reactions in combi-
nation with PC. These results are similar to
previous research findings by Monzó et al
(2003), Tantawy et al (2012), Horiguchi et
al (2011) and Naamane et al (2016).
Microscopic assessment of
hydrated PC/ISSA pastes
Figure 12 shows selected micrographs
of hydrating PC and PC/ISSA pastes.
Hydration products deposited on the
surfaces of particles were identified by
their morphology and chemical composi-
tion based on EDS analysis. At six hours
after mixing, the PC paste (Figure 12(a))
showed typical deposits of outer calcium-
silicate-hydrate (CSH), calcium hydroxide or
Portlandite (CH) and needles of ettringite,
attached to the PC grains. In the case of the
ISSA pastes, a potassium gel deposit was
observed in the 8F and 8A paste samples
at six hours after mixing (see Figure 12(b)).
This gel has a solid morphology and appears
to have a retarding effect on hydration of the
ISSA (Odler & Wonnemann 1983), as there
was little evidence of hydration products
on ISSA surfaces in these samples at this
early stage of hydration. Importantly, the gel
was not observed in the 8Q ISSA-blended
cement paste and there was some evidence
of fibrous hydration products on ISSA
particles in the 8Q sample, which may be
44
(a) (b) (c)
Figure 11 M
 icrographs of ISSA incinerated at 800°C, then (a) furnace-cooled, (b) air-cooled and
(c) quenched in water
Figure 12 S amples of SEM images of hydrating binder pastes: (a) PC paste at six hours, (b) paste
with 30% 8F ISSA at six hours, (c) paste with 30% 8Q ISSA at seven days
40
35
30
25
Flow %
20
15
10
5
0
PC PC/FA PC/7F PC/7A PC/7Q PC/8F PC/8A PC/8Q PC/9F PC/9A PC/9Q
Mortar binder type
Figure 13 E ffect of ISSA materials on workability of mortars when included at 30% replacement
of PC in the binder, compared with equivalent replacement of FA and plain PC mortars
linked to early hydration of ISSA materials blend caused a slight decrease in the flow
in the paste. of the mortar, compared to that of the plain
At seven days after mixing, there was PC mortar, all the 30% ISSA blends result-
little evidence of the potassium gel in any ed in a significant reduction of workability.
of the ISSA-blended pastes. All the samples This reduction ranged from 11 percentage
showed some evidence of hydration prod- points for the 9Q ISSA to 18 percent-
ucts on the ISSA particles (see Figure 12(c)), age points for the 7F ISSA. Again, these
but it was evident that the post-incineration results are similar to that noted by other
cooling rate had an influence on the extent researchers (Vouk et al 2016; Naamane et
of hydration. The quenched ISSA sample al 2016; Pan et al 2003; Monzó et al 1996).
generally showed a greater degree of hydra- Importantly, the quenched ISSA samples
tion than the furnace-cooled or air-cooled generally showed the smallest reduction
ISSA samples. in workability compared with that of the
furnace-cooled and air-cooled samples.
Workability of mortars The reduction in mortar workability is
containing ISSA attributed mainly to the irregular morphol-
The results of workability measurements ogy of ISSA particles. However, the possible
conducted on each of the mortars prepared increased water absorption of the ISSA
are shown in Figure 13. While the 30% FA materials may also be a factor contributing
Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
to this reduced workability. Vouk et al
(2016) and Naamane et al (2016) report that 60
an increased water absorption of the ISSA
simultaneously resulted in longer setting 50

Compressive strength (MPa)

times, and reduced the mortar workability
due to the high amounts of phosphorus 40

pentoxide (P2O5) and sulphur trioxide

30
(SO3) found in the ISSA.

20
Compressive strength test results
Figure 14 shows the compressive strength
10
development of all mortar samples after 3,
7 and 28 days, each result being the average
0
of three cube strength tests. The results PC PC/FA PC/7F PC/7A PC/7Q PC/8F PC/8A PC/8Q PC/9F PC/9A PC/9Q
show that, at a PC replacement level of 30%:
QQ All the ISSA materials used as an SCM
caused a reduction in the measured
compressive strength at all ages up to 28 Figure 14 Compressive strength test results of mortar samples after 3, 7 and 28 days of hydration
days, relative to that of the mortar made
with plain PC. The mortar compressive strength
QQ For the mortars with ISSA materials results shown in Figure 14 support the
incinerated at 700oC and at 800oC, the observations made in the SEM study of the
compressive strengths were significantly hydrated pastes (Figure 12) that the ISSA
lower than that of the mortar with FA materials do show evidence of pozzolanic
as SCM. However, the mortars with activity when used as an SCM with PC. For
ISSA incinerated at 900oC showed the ISSA incinerated at 900oC, the 28-day
compressive strengths that were similar compressive strength performance was
to that of the FA mortar. similar to that of the FA mortar.
QQ The incineration temperature of the Of course, a factor not accounted for in
ISSA has a significant effect on mortar this analysis was the fineness of grinding
compressive strength. A higher cooling of the ISSA. Halliday et al (2012) showed
rate after incineration generally has a that coarser ISSA tends to contribute to
positive effect on compressive strength. the reduction in compressive strength.
For the ISSA materials incinerated at The particle size distributions shown in
800oC and at 900oC, the quenched sam- Figure 4 indicate that the ISSA used in this
ples showed marginally higher strengths study was coarser than both the PC and
than the furnace-cooled and air-cooled the FA. It is therefore possible that higher
materials. This increased performance of
the quenched ISSA may be explained by Table 4 R
 ates of early mortar strength gain
the fact that rapid cooling of the ash sup- relative to the 28-day strength of the
presses the formation of crystalline silica, binder combinations assessed
resulting in a higher proportion of amor- Mortar 3-day to
phous silica which is more favourable for binder 28-day
the pozzolanic reactions in an SCM. type strength (%)
QQ Table 4 shows an assessment of the PC 62.6
rates of early strength development for
PC/FA 53.3
the mortars assessed. It is clear that
the ISSA materials have significantly PC/7F 43.4
reduced the rate of strength gain at
PC/7A 47.8
three days, to a greater extent than that
caused by the addition of FA. However, PC/7Q 43.2
there is some acceleration in strength
PC/8F 46.6
gain thereafter and, for the period 7
to 28 days, the rate of strength gain is PC/8A 44.2
similar to that obtained with the PC/‌FA PC/8Q 42.3
mortar samples. Importantly, all the
PC/9F 41.5
blended cement mortars showed lower
rates of strength gain between 7 and PC/9A 44.7
28 days when compared with the plain
PC/9Q 49.7
PC mortar.
Mortar binder type
3 days 7 days 28 days
levels of pozzolanicity may be realised with
more finely ground ISSA.
Manjunatha (2017) found a lower
pozzolanic activity for ISSA than for FA.
However, the present study has shown that
under suitable pyro-processing regimes,
ISSA can show similar strength perfor-
mance as FA when used as an SCM. The
differences in these findings may well be
related to the chemical nature of the FA
and the ISSA, as well as the physical pro-
cessing of the ISSA after cooling.
Conclusions
This study was aimed at considering the
possibility that incinerated sewage sludge
ash could be used as a supplementary
cementitious material in combination with
Portland cement. The main conclusions to
be drawn from the results are:
QQ The SEM assessment of paste samples,
7-day to together with the mortar compressive
28-day strength results, indicates that the
strength (%) ISSA used in this study has pozzolanic
86.3 capacity and may be used as an SCM in
cement-based materials.
75.0
QQ DSS and ISSA consists mainly of Si, Ca,
79.5 Al, Fe, Mg, Na and K, the important ele-
ments in cementitious and pozzolanic
77.9
materials used in concrete. The concen-
77.1 trations of these elements increase with
incineration and produces an ISSA with
79.2
a chemical profile similar to that of FA
75.4 in terms of oxide composition.
73.3 QQ Heavy elements that may also be toxic
such as Zn, Pb, Cu, Ti, Cr, Mn, V and
70.9
Ni are present as minor constituents.
71.7 The amounts of these heavy elements
are reduced by incineration and there
72.2
may be environmental benefits to

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 45
 incorporating such elements in concrete
rather than in land disposal systems.
However, removal of these elements
from the flue gasses is also important to
avoid release into the atmosphere.
QQ Both the maximum temperature of
incineration and the subsequent cooling
rate of ISSA have a significant influ-
ence on the workability, hydration and
strength development of pastes and
mortars made with blends of PC and
ISSA. In general, increasing the incin-
eration temperature of ISSA from 700oC
to 900oC, together with rapid cooling
by quenching in water, produced better
early hydration, better workability and
higher mortar strength when the ISSA
is used as an SCM at 30% replacement
of PC. The ISSA incinerated at 900oC
and quenched in water produced 28-day
mortar strengths that are similar to that
obtained with mortars made with FA at
the same replacement level.
QQ When used as a PC replacement in
paste samples, the ISSA materials that
were incinerated at 700oC and 800oC
showed deposits of potassium gel after
six hours of hydration that appear to
inhibit early hydration of the ISSA.
This gel was not evident at later ages of
hydration. In the case of the ISSA incin-
erated at 900oC, the gel was not noted
at any stage of assessment of hydration.
This aspect of the hydration behaviour
of ISSA requires further assessment
to understand its implications for
the use of these materials in cement
and concrete.
Acknowledgements
The authors are thankful to SASOL for
the financial support provided to complete
the work, and to AfriSam (Pty) Ltd for
their assistance with some of the chemical
analyses reported in this work.
References
ASTM (American Society for Testing and Materials)
2000. ASTM D 854-00. Standard Test Methods for
Specific Gravity of Soil Solids by Water Pycnometer.
West Conshohocken, PA: ASTM International.
ASTM 2007. ASTM C1437-07. Standard Test Method
for Flow of Hydraulic Cement Mortar. West
Conshohocken, PA: ASTM International.
Ballim, Y & Graham, P C 2009. The effects of
supplementary cementing materials in modifying
the heat of hydration of concrete. Materials and
Structures, 42(6): 803‒811.
46
Chen, L & Lin, D F 2009. Stabilization treatment of
soft subgrade soil by sewage sludge ash and cement.
Journal of Hazardous Materials, 162(1): 321‒327.
Coutand, M, Cyr, M & Clastres, P 2006. Use of sewage
sludge ash as mineral admixture in mortars.
Proceedings of the Institution of Civil Engineers –
Construction Materials, 159(4): 153‒162.
Cyr, M, Coutand, M & Clastres, P 2007. Technological
and environmental behavior of sewage sludge ash
(SSA) in cement-based materials. Cement and
Concrete Research, 37(8): 1278‒1289.
Dyer, T D, Halliday, J E & Dhir, R K 2011. Hydration
chemistry of sewage sludge ash used as a
cement component. Journal of Materials in Civil
Engineering, 23(5): 648‒655.
Fleischman, N D, Botha, M F & Germanis, J P 2014.
Sewage sludge and biomass incineration in South
Africa using a fluidized-bed reactor. South African
Journal of Chemical Engineering, 19(3): 35‒47.
Garcés, P, Carrión, M P, García-Alcocel, E, Payá, J,
Monzó, J & Borrachero, M V 2008. Mechanical and
physical properties of cement blended with sewage
sludge ash. Waste Management, 28(12): 2495‒2502.
Gomes, S D C, Zhou, J L, Li, W & Long, G 2019.
Progress in manufacture and properties of
construction materials incorporating water
treatment sludge: A review. Resources. Conservation
and Recycling, 145: 148‒159.
Grobelak, A, Czerwińska, K & Murtaś, A 2019. General
Considerations on Sludge Disposal, Industrial
and Municipal Sludge. Oxford, UK: Butterworth-
Heinemann, pp 135–153.
Halliday, J E, Jones, M R, Dyer, T D & Dhir, R K 2012.
Potential use of UK sewage sludge ash in cement-
based concrete. Proceedings of the Institution of
Civil Engineers ‒ Waste and Resource Management,
165(2): 57‒66.
Horiguchi, T, Fujita, R & Shimura, K 2011.
Applicability of controlled low-strength materials
with incinerated sewage sludge ash and crushed-
stone powder. Journal of Materials in Civil
Engineering, 23(6): 767‒771.
Lewis, D L & Gattie, D K 2002. Pathogen risks from
applying sewage sludge to land. Environmental
Science & Technology, 36(13): 286A‒293A.
Manjunatha, M 2017. Use of sewage sludge ash and fly
ash in construction. International Journal of Latest
Engineering Research and Applications, 8(2): 102‒110.
Massazza, F 1998. Pozzolana and pozzolanic cements.
Chapter 10. In Hewlett, P C (Ed.), Lea’s Chemistry of
Cement and Concrete, 4th ed. London: Elsevier.
Monzó, J, Paya, J, Borrachero, M V & Córcoles, A 1996.
Use of sewage sludge ash (SSA)-cement admixtures
in mortars. Cement and Concrete Research,
26(9): 1389‒1398.
Monzó, J, Paya, J, Borrachero, M V, Bellver, A &
Peris‑Mora, E 1997. Study of Cement-Based
Mortars Containing Spanish Ground Sewage Sludge
Ash. Studies in Environmental Science Vol 7.
Amsterdam, Netherlands: Elsevier, pp 349–354.
Volume 64 Number 1 March 2022
Monzó, J, Payá, J, Borrachero, M V & Girbés, I 2003.
Reuse of sewage sludge ashes (SSA) in cement
mixtures: The effect of SSA on the workability
of cement mortars. Waste Management,
23(4): 373‒381.
Naamane, S, Rais, Z & Taleb, M 2016. The effectiveness
of the incineration of sewage sludge on the evolution
of physicochemical and mechanical properties
of Portland cement. Construction and Building
Materials, 112: 783‒789.
National Research Council (US) 2002. Biosolids
Applied to Land: Advancing Standards and
Practices. Washington, DC: National Academies
Press.
Odler, I & Wonnemann, R 1983. Effect of alkalies on
Portland cement hydration. Part II. Alkalies present
in form of sulphates. Cement and Concrete Research,
13(6): 771‒777.
Osbaeck, B & Johansen, V 1989. Particle size
distribution and the rate of strength development of
Portland cement. Journal of the American Ceramic
Society, 72(2): 197‒201.
Pan, S C, Tseng, D H, Lee, C C & Lee, C 2003.
Influence of the fineness of sewage sludge ash on the
mortar properties. Cement and Concrete Research,
33(11): 1749‒1754.
Pavlík, Z, Fořt, J, Záleská, M, Pavlíková, M, Trník,
A, Medved, I et al 2016. Energy-efficient thermal
treatment of sewage sludge for its application in
blended cements. Journal of Cleaner Production,
112: 409–419.
Payá, J, Monzó, J, Borrachero, M V & Soriano, L
2019. Sewage sludge ash. In De Brito, J & Afrela,
D (Eds.), New Trends in Eco-efficient and Recycled
Concrete. Sawston, Cambridge, UK: Woodhead
Publishing Publishing Series in Civil and Structural
Engineering, pp 121–152.
Rakhimov, R Z, Rakhimova, N R & Ojovan, M I 2015.
Alternative binders. Cementitious Materials for
Nuclear Waste Immobilization, 4: 93.
Scrivener, K L, John, V M & Gartner, E M 2017. Eco-
efficient cements: Potential, economically viable
solutions for a low-CO2 , cement-based materials
industry. United Nations Environment Program
report.
Tantawy, M A, El-Roudi, A M, Abdalla, E M &
Abdelzaher, M A 2012. Evaluation of the pozzolanic
activity of sewage sludge ash. Kerala, India: ISRO
Chemical Engineering (Indian Space Research
Organisation).
Tay, J H 1987. Sludge ash as filler for Portland cement
concrete. Journal of Environmental Engineering,
113(2): 345‒351.
Tenza-Abril, A J, Saval, J M & Cuenca, A 2015.
Using sewage-sludge ash as filler in bituminous
mixes. Journal of Materials in Civil Engineering,
27(4): 04014141.
Vimonsatit, V, Chindaprasirt, P, Ruangsiriyakul, S &
Sata, V 2015. Influence of fly ash fineness on water
requirement and shrinkage of blended cement
Journal of the South African Institution of Civil Engineering
 mortars. Engineering and Applied Science Research,
42(4): 311‒316.
Vouk, D, Serdar, M, Nakić, D & Anić-Vučinić, A 2016.
Use of sludge generated at WWTP in the production
of cement mortar and concrete. Građevinar,
68(03):199‒210.
Vouk, D, Serdar, M & Vučinić, AA 2017. Use of activated sludge system for treating coal gasification
incinerated sewage sludge ash in cement mortars: wastewater. Chinese Journal of Chemical
Case study in Croatia. TehnickiVjesnik/Technical Engineering, 17(1): 167‒170.
Gazette, 24(1).
Zhang, W, Pinhua, R A O, Zhang, H & Jingli, X U
2009. The role of diatomite particles in the

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 47
TECHNICAL PAPER Evaluation of the seismic
Journal of the South African
Institution of Civil Engineering response of a reinforced
ISSN 1021-2019
Vol 64 No 1, March 2022, Pages 48–62, Paper 0814 concrete footing with
SHANE HOSSELL (Pr Eng) is a structural stub column to increasing
peak ground acceleration
engineer at Zutari, Pretoria. He obtained
BEng and MEng degrees in Civil Engineering
at the University of Pretoria, and is currently

using pseudo-dynamic
working on the analysis and design of
bridges located in Australia and New
Zealand. His interests are in structural

experimentation
analysis and structural dynamics.

Contact details:
Department of Civil Engineering
University of Pretoria
Pretoria 0002
South Africa S M Hossell, C P Roth
T: +27 82 806 1136
E: shanehossell@gmail.com

PROF CHRIS ROTH (Pr Eng, FSAICE) is

The pseudo-dynamic experimentation technique was investigated to evaluate the damage
Associate Professor in Civil Engineering at occurring in a reinforced concrete footing with stub column due to the overall response of a
the University of Pretoria, working in the linear elastic two-storey, two-bay moment-resisting steel frame structure that is subjected to
discipline of structural engineering. He
an earthquake excitation with increasing peak ground acceleration. The implicit Newmark’s
started his career in consulting engineering
before joining the University of Pretoria, and method with static condensation was utilised in the present study to solve the governing
is the current chairman of SANS Technical equation of motion of the multi-degree-of-freedom system. Five pseudo-dynamic experiments
Committee 98/Subcommittee 1 on Basis of were performed by scaling the El Centro ground motion record, which occurred in California
Design and Actions (including Earthquake Design). He obtained a BEng
on 18 May 1940, to produce peak ground accelerations that ranged between 0.34 g and 2 g.
degree in Civil Engineering at the University of Stellenbosch, and an MS
and PhD at Cornell University. His interests are in structural reliability and All the laboratory experiments were undertaken under a constant axial load for the duration
structural analysis. of the applied earthquake excitation, and utilised Rayleigh damping to model the energy loss
Contact details: within the overall structure. The pseudo-dynamic method provides a reliable method to relate
Department of Civil Engineering damage suffered by the stub column due to the overall structure’s response to the applied
University of Pretoria earthquake excitation. The method enables the structural capacity and failure mechanisms
Pretoria 0002
South Africa
of the reinforced concrete stub column to be observed in relation to the seismic demand.
T: +27 12 420 2185 The hysteretic response of the stub columns and energy dissipation characteristics were
E: chris.roth@up.ac.za determined, and it was shown that the yield strength of the longitudinal reinforcement within
the stub column has a significant impact on the maximum shear capacity and damage incurred
by the stub column. The damage is more pronounced with an increase in the number of cycles
of vibration, particularly at displacements that exceed the yield strength of the reinforcement.
An increase in the hysteretic energy dissipated by the reinforced concrete stub column results
in a concomitant increase in the observed damage to the stub column in the form of concrete
cracking, reinforcement yielding and spalling of the concrete.

Introduction system comprising computer software

Keywords: pseudo-dynamic experimentation, reinforced concrete
stub column, seismic performance evaluation,
hysteretic curves, Newmark’s implicit numerical method,
El Centro earthquake
The pseudo-dynamic testing technique is a
computer-controlled experimental method
whereby the dynamic behaviour of the
structure is mathematically calculated on
a computer and the resultant displacement
is statically applied to the physical test
specimen of the structure at a common
degree of freedom between the numerical
model and the physical model using servo-
controlled actuators in an online procedure
(Mosalam et al 1997). The pseudo-dynamic
testing technique uses the same equipment
as conventional quasi-static tests; however,
the analysis is controlled by a closed-loop
that is integrated and runs in tandem with
the quasi-static experiment. The method
utilises well-established step-by-step time
integration methods and utilises the same
numerical approach generally undertaken
in nonlinear structural dynamics; however,
the structural restoring force is based on
experimental feedback from load cells as
opposed to an idealised material model
(Wang et al 2006). The force obtained from
the load cell, due to the calculated displace-
ment from the numerical model, is fed back
into the computational model and used
in successive iterations to determine the

Hossell SM, Roth CP. Evaluation of the seismic response of a reinforced concrete footing with stub column to increasing peak ground acceleration using
48 pseudo-dynamic experimentation. J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #0814, 15 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64no1a5
new displacement. Historically, reinforced
concrete stub columns have been experi-
mentally tested using both quasi-static
monotonic load tests and cyclic load tests
to determine the performance of the struc-
ture; however, the experiments are not able
to relate the progressive damage incurred
to the reinforced concrete stub column
for the duration of a realistic earthquake
excitation (Mosalam et al 1998). Therefore,
the pseudo-dynamic experimental method
enables the cumulative damage to be
related to earthquake intensity (Mosalam et
al 1998; Takanashi et al 1975).
Background
The performance of structures during
an earthquake of given intensity dictates
the extent of damage and loss of life that
become associated with the earthquake
event. Quantifying the level of damage
within a structure that has occurred during
an earthquake is traditionally undertaken
post-earthquake using statistical methods.
However, this method is not suitable in
areas with moderate seismicity, as insuffi-
cient data is available to calibrate structural
damage to an earthquake intensity param-
eter. Southern Africa is characterised as
a region of moderate seismicity, and due
to the limited network of accelerometers
within South Africa, a detailed statisti-
cal analysis of the level of damage that
could occur within structures during
future earthquakes has been prevented
(Brandt 2011).
Quantifying the level of damage
incurred by a structure due to increasing
earthquake intensity is a complex task.
Typically, structural components are
evaluated using quasi-static methods to
determine the response due to increasing
load and to determine the ultimate load-
carrying capacity of the member. However,
the slow rate of the load applied onto
the structure results in the inertia of the
structure not being considered, resulting in
the response of the structure being inde-
pendent of the applied earthquake loading.
To relate earthquake intensity to damage,
shake-table testing or pseudo-dynamic
testing provides a more accurate damage
correlation with earthquake intensity.
Shake-table tests provide the most realistic
means to evaluate damage at various
intensities, as they account for the inertial
effects and the time and frequency content
of the ground motion. However, shake
tables are very expensive and difficult to
evaluate large-scale multi-storey structures
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
Reinforced
concrete
footing
under
investigation
ag t = i∆t
Earthquake loading
t
Figure 1 Conceptual model of the overall structure
for a range of earthquake intensities.
Advancements in computer software, the
increase in the resolution of the control
systems and data acquisition systems, and
the ability of pseudo-dynamic tests to
incorporate the dynamic characteristics
of a structure have made the method a
feasible alternative to shake-table tests to
evaluate the performance of a structure at
various earthquake intensities. This could
aid in the development of fragility curves
and in determining the risk of damage at
various earthquake intensities.
Objective of the study
The primary objective of the study was to
demonstrate the application of the pseudo-
dynamic experimental method to relate
the resultant structural damage, hysteretic
response and energy dissipation capacity
of a reinforced concrete footing and stub
column under a constant axial load that
forms part of a two-bay two-storey moment-
resisting steel frame structure, to increasing
peak ground acceleration (PGA). Figure 1
shows the hypothetical structure that was
considered in this research, and the position
of the reinforced concrete footing and stub
column under investigation. The feasibility
of using the pseudo-dynamic method with
Newmark’s implicit time-stepping numeri-
cal method with static condensation was
investigated by analysing a single member
of a structure that forms part of an overall
structural system that remains linearly
elastic for the duration of the experiment.
The study also investigated the maximum
peak ground acceleration that the member
Brick infill walls
Structural steel column
Reinforced concrete slab
Structural steel beam
Reinforced concrete footing
could encounter before failure, and there-
fore the experiments were undertaken at
increasing amplitudes.
The footing shown in Figure 1 was
assumed to be fixed against rotation, and
the steel structure above was assumed
to be connected to the footing and stub
column by a pin joint. The effect is that the
footing and stub column assembly provides
an infinitely stiff vertical support to the
steel structure, and a nonlinear spring form
of horizontal support.
Limitations of the study
The susceptibility of damage to a rein-
forced concrete footing and stub column
during an earthquake is dependent on
several factors that relate to the type of
structure placed on the footing and stub
column, the ground conditions, the bound-
ary conditions, and the type of earthquake
excitation to which the structure is sub-
jected. The size of the reinforced concrete
footing and stub column that could be
evaluated and the magnitude of the applied
axial load were limited by the capacity of
the press frame that was available in the
laboratory. The study focused on evaluat-
ing the pseudo-dynamic testing method
using only the Newmark implicit numerical
time integration method, and only the El
Centro ground motion record was used in
the analysis, with the peak ground accel-
eration of the earthquake excitation being
the only variable. The reinforced concrete
footing and stub column was subjected
to a constant axial load for the duration
of the earthquake record, and therefore
49
the response of the stub column due to a
varying axial load was not investigated.
Only a reinforced concrete footing and
stub column of a single design that satisfied
the minimum reinforcement requirements
contained in SANS 10100-1:2000 (SANS
2000) was considered in this research. The
two-dimensional frame structure, which
was placed on the footing and stub column,
remained linearly elastic for the duration
of the earthquake record, and energy loss
within the overall frame structure was
treated using Rayleigh damping with a
damping ratio of 5%, as is typically used in
design codes (SANS 2017; Chopra 2012).
The capacity of structural members and
their connections within the overall frame
structure were not considered during the
analysis, and therefore the formation of
plastic hinges and the resulting loss of stiff-
ness within the overall structure were not
considered when determining the response
of the reinforced concrete footing and stub
column to the applied earthquake load.
Relevant literature
Pseudo-dynamic testing originated
approximately forty years ago as an alterna-
tive to shake-table testing, with Takanashi
et al (1975) seen as one of the pioneers of
the testing method (Kurt 2010). Pseudo-
dynamic testing is especially efficient
when having to test structures that are
too heavy or too large to be practically
tested on available shake tables (Thewalt &
Mahin 1987). Pioneering work in pseudo-
dynamic testing was done by Hakuno in
1969 (as cited by Takanashi et al 1975)
where he tested cantilever beams using an
online system that comprised an analogue
computer and an electromagnetic actuator.
However, the results produced by the test
were rather poor due to the limitations of
the available hardware. Following on the
work that was done on cantilever beams,
Takanashi et al (1975) did substantial work
in establishing the pseudo-dynamic tech-
nique by replacing the analogue computer
with a more accurate digital computer.
Modifications done by Takanashi et al
(1975) enabled the procedure to not have
to operate in real time, thus producing a
method that could be subjected to slow
loading and pausing. The first pseudo-
dynamic tests were restricted to planar test
specimens that were subjected to a single
horizontal component of base excitation
(Takanashi et al 1975; Shing & Mahin
1984; Takanashi & Nakashima 1987).
50
However, the method can be easily adapted
to analyse the three-dimensional response
of structures with several components of
base excitation (Thewalt & Mahin 1987).
Multi-degree-of-freedom testing was done
by Chang (2009) whereby he subjected a
one-storey frame to bidirectional loading.
Mosalam et al (1998) investigated the
response of masonry infill frames using
the pseudo-dynamic method and indicated
that the pseudo-dynamic method provides
an acceptable approximation of the dynam-
ic response of a structure that is subjected
to earthquake excitation.
Formulation of the
pseudo-dynamic method
The pseudo-dynamic method is formulated
from the time-discretised equation of
motion for each time step “i” as shown
by Equation 1. During pseudo-dynamic
experimentation, Equation 1 is solved using
numerical methods in a stepwise procedure
with the restoring force Fs = [Ks(t)]{u(t)},
measured directly from the test specimen
using a load cell (Kurt 2010). Therefore, the
value of [Ks(t)] is calculated from the meas-
ured value Fs force obtained from the load
cell due to the displacement {u(t)} applied
by the linear hydraulic actuator to the test
specimen and calculated in the numerical
model of the overall frame structure. The
solution of the second order differential
equation can be solved using either implicit
or explicit numerical methods to solve the
displacements at each time step.
�����������������������������
[M]{ü(t)} + [C]{u̇(t)} + [K]{u(t)} +
Linear
component
�����
[K ��� – �����
s(t)]{u(t)} = {P} [M]{I}üg (t)(1)
Non-linear Static Earthquake
component load loading
Where:
[M]: Lumped mass matrix of the
frame structure including the
mass of the reinforced con-
crete footing and stub column
[C]: Rayleigh damping matrix for
the elastic part of the frame
structure
[K]: Linear elastic stiffness matrix
of the frames structure
{ü(t)}, {u̇(t)}, Acceleration, velocity and
{u(t)}: displacement vector
respectively
[Ks(t)]: Lateral non-linear spring stiff-
ness matrix of the reinforced
Volume 64 Number 1 March 2022
concrete footing and stub
column
{I}: Influence vector that accounts
for the horizontal direction of
the earthquake loading
üg (t): Ground acceleration.
The preference in the past has traditionally
been towards explicit numerical methods
because of the disinclination to numeri-
cal iteration at each time step. However,
advancements in computational power and
the increase in the resolution of computers
have made using implicit numerical meth-
ods more favourable due to the superior
stability properties they provide (Mosalam
et al 1997). Newmark’s implicit method has
been widely adopted in finite element ana­
lysis software to solve nonlinear problems
as it can be unconditionally stable for any
time increment (Chopra 2012). Therefore,
the selection of the time increment only
influences the accuracy of the solution and
not the stability.
EXPERIMENTAL PROCEDURE
Five pseudo-dynamic experiments were
conducted on the reinforced concrete
footings and stub columns at increasing
peak ground accelerations (PGA) obtained
by amplifying the El Centro earthquake
to correlate damage with increasing peak
ground acceleration. The PGA considered
during this study comprised 0.34 g, 0.68 g,
0.78 g, 1 g and 2 g. The testing procedure
required the programming of the implicit
Newmark numerical algorithm into the
HBM Catman software (HBM 2016)
using the Visual Basic scripting language,
and the software was integrated with the
hardware upon completion of the scripting.
The displacement transducers and load
cells were calibrated before testing, and
the integration between the hardware and
the software was tested to ensure that the
servo-controller accurately interpreted the
calculated displacement.
Each of the experiments started by
initialising the actuators and ensuring
the horizontal load on the stub column
was equal to zero. The axial load was
then applied to the stub column, and
once the axial load reached 300 kN, the
horizontal actuator was initialised on the
servo-controller. The initial calculation
within the software ensured that all the
instruments were zeroed, and the static
analysis was first performed to determine
the initial load state of the structure before
Journal of the South African Institution of Civil Engineering
 0.4

0.3

0.2
Ground acceleration (g)

0.1

–0.1

–0.2

–0.3
0 5 10 15 20 25 30 35 40 45 50 55
Time (s)

Figure 2 Acceleration record of the El Centro earthquake

commencing the pseudo-dynamic analysis.

Time-stepping and iteration commenced 300 300
upon completion of the initial calcula-
tions whereby the amplified El Centro
earthquake record was applied to the 4 × M16 class 8.8 holding down bolts
overall structure.
The initial horizontal stiffness of the
1 000

7 × Y10 stirrups shear reinforcement

reinforced concrete footing and stub
column was required before the pseudo-
dynamic analysis could commence. The 30 MPa concrete stub column
initial stiffness of the footing and stub
7 × Y12 tensile reinforcement
column also enabled the elastic natural
period of vibration of the structure to be
determined, which was used to calcu-
late the damping matrix for the overall 30 MPa concrete
300

foundation pad slab

frame structure.

Earthquake loading on the structure

The El Centro earthquake, S00E of the 1 10
event at Imperial Valley, California, on 0

18 May 1940, was selected as the input 700

ground motion and is shown in Figure 2.

The El Centro ground motion record
represents a strong ground shaking
and is commonly used for shake-table
tests and pseudo-dynamic experiments
(Mosalam et al 1998). For each of the
pseudo-dynamic tests undertaken in this
study, the peak ground acceleration was
increased by scaling the amplitude of the
El Centro earthquake.
Test specimen
The reinforced concrete footing and stub
column that were used in the research are
shown in Figure 3. The reinforced concrete
stub column only required minimal rein-
forcement to resist the applied gravity and
wind loads that were used to design the
Figure 3 T hree-dimensional visualisation of the reinforced concrete footing and stub column
(units in mm)
structure. The characteristic design yield
stress of 450 MPa and a Young’s Modulus
of 200 GPa were used for the design of the
reinforcement, and a characteristic com-
pressive strength of 30 MPa was used for
the concrete.
Application of the pseudo-
dynamic method using
Newmark’s implicit method
Figure 4 shows the discretised numeri-
cal model used in the pseudo-dynamic
analysis procedure in this research. The
frame structure was discretised into 28
degrees of freedom with the boundary
conditions comprising two pin supports at
the external columns, and with the internal
stub column’s lateral degree of freedom
idealised as a single degree of freedom
consisting of a constant lumped mass that
was supported laterally by a massless non-
linear spring of stiffness ks(t). The degree of
freedom {us = u1} coupled the calculation
cycle with the laboratory test specimen.
The horizontal displacement {u1} was taken
at the top of the stub column, as shown

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 51
in Figure 4, where the steel column of the
frame structure connects to the footing
and stub column using a base plate and 7 8 9
θ20 θ21 θ22
holding-down bolts. u8 9 u10 10 u12

Loading cycle – physical test setup Node

A conceptual model of the experimental v9 v11 v13
test setup is shown in Figure 5. The test 2 4 6
setup comprised two actuators and two
4 θ17 5 θ18 6 θ19
load cells, with the vertical actuator being
u2 7 u4 8 u6
used to simulate the constant axial load
applied by the structure on the stub col-
umn, which corresponds to the degree of v3 Structural v5 v7
elements 3
freedom v 25 shown in Figure 4. The hori-
1 5
zontal actuator was used to simulate the θ15
varying horizontal shear load on the stub 1 u1 3
θ14 θ16
u23 = 0 u28 = 0 u26 = 0
column due to the applied earthquake exci- 2
10 ks
tation that was calculated from the compu-
tational model of the frame structure.
The displacement us calculated in the
computer model was applied directly to the v 24 = 0 v 25 = 0 v 27 = 0

stub column using the horizontal actuator

Equivalent mass
with the corresponding force Fs measured spring system
using the horizontal load cell [LC1]. The
horizontal force was subsequently fed back
(a) Numerical model
into the numerical model at which time the
numerical model computation continued.
Figure 6 shows the hardware and software
Faxial Faxial
that were used to integrate the calculation
cycle with the loading cycle on the test
specimen. Two control loops were used in u1(t) = us(t)
the experimental tests, with the first con- u1(t) = us(t)
Fs(t)
trol loop being used under force control to M M
maintain the vertical axial load on the rein-
forced concrete stub column. The second
control loop was used under displacement
control to apply the horizontal displace- M ks(t)
ment calculated in the numerical model
onto the reinforced concrete stub column
θ=0
at each iteration within each time step. A θ=0
linear ramp function was used to apply the
load incrementally and delay the rate of
lateral load applied onto the stub column.
The objective of the ramp function was
to mitigate dynamic effects during the * Rotation of footing is not considered
analysis due to large changes in calculated
displacements between iterations.
(b) Physical model
Figure 6 also shows a third path that
comprised additional instruments attached
to the experimental test setup. Strain Figure 4 Pseudo-dynamic numerical model and physical model
gauges were attached to two of the four
longitudinal reinforcing bars. The strain Calculation cycle numerical model mass within the overall frame structure. The
gauges were applied at the base of the The nonlinear Newmark’s implicit method, unknown displacements and forces within
column where the maximum moment was using the average acceleration method, the structure using the initial state of the
expected. Four additional linear displace- was used to perform the pseudo-dynamic structure are solved at each time step:
ment transducers were placed at equal experiments, and Figure 7 shows the pseudo-
increments over the length of the column dynamic algorithm used to run the analysis. i = 1, 2, 3 … n with t = i∆t
with the aim to validate the results with The algorithm was adapted to incorporate
the internal displacement transducer in the static condensation to eliminate the rota- With n being the total number of time
horizontal actuator. tional free degrees of freedom with zero steps in the earthquake record, and t being

52 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 the time in seconds at any point within
the analysis.
Vertical press frame The average acceleration method
Horizontal load cell [LC1] produces an unconditionally stable solu-
Dywidag bar tion, and therefore the selection of the
Horizontal actuator Vertical actuator time step only influenced the accuracy of
the solution and not the stability of the
Vertical load cell [LC2]
Displacement solution. For each time step “i” within
transducer Actuators adaptor
connection the earthquake record, iteration “ j” was
Footing required due to the implicit relationship
Reaction frame
LVDT frame
between the restoring force matrix ( f*s)‌i
and the unknown displacement of the
Steel pins
structure at (u)i as shown in Step 15 in
Press base connection
Figure 7. Therefore, the stiffness of the
footing and stub column must be assumed
before the calculation at each time step can
be initialised. The selection of the initial
lateral stiffness of the footing and stub
Test column had to be considered carefully,
Overturning floor
restraint frame
because assuming a stiffness that was lower
than the true elastic stiffness of the footing
Sliding restraint frame
and stub column would produce premature
Reaction frame test floor connection damage to the stub column at the start of
each time step (Pegan & Pinto 2000). Also,
Figure 5 Conceptual experimental test setup selecting a stiffness that is lower than the

On-line

Numerical model
M M M
ag
Ramp fuction t = i∆t
Digital to analog
converter (DAC) {ds} M M M

i t
(HBM PMX system) Fs
Mf us
(16 bit DAC) i–1
ks
Output
constant
Numerical integration using axial load
Newmark’s non-linear algorithm

{Fs}
Zwick-Roell K7500
servo-controller Analog to digital converter (ADC)
(displacement control) HBM PMX / QuantumX
24 bit ADC Digital to
analog
Displacement Displacement converter
control servo- transducer Servo-valve
hydraulic {dinternal}
control loop
Vertical Load sontrol
actuator servo- Servo-controller
Servo-valve hydraulic (force control)
Horizontal actuator Load Load cell control loop
cell
Load cell
reading {Faxial}

Digital to
LVDT {dexternal} analog
converter

Strain gauges

Figure 6 Schematic illustration of external components of the pseudo-dynamic experiment

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 53
 1. Initial conditions
1.1 State determination {fs}0 and {u}0 – initial force and displacement of the structure
1.2 Select ∆t – time increment
1.3 Predetermined mass matrix [Mtt] and damping matrix [Ctt]
1.3 u̇0 = 0 and ü0 = 0 (initial structure is stationary)

[α1] = ⎫⎪
1 ⎫ ⎫ γ ⎫ ⎫ 1 ⎫ ⎫γ ⎫ ⎫1 ⎫ ⎫γ ⎫
Newmark’s coefficients average acceleration method γ = ½, β = ¼

⎪[Mtt] + ⎪β∆t⎪[Ctt]    [α2] = ⎪β∆t⎪[Mtt] + ⎪ β – 1⎪[Ctt]   [α3] = ⎪2β – 1⎪[Mtt] + ∆t⎪2β – 1⎪[Ctt]

⎭β(∆t)2⎭ ⎭ ⎭ ⎭ ⎭ ⎭ ⎭ ⎭ ⎭ ⎭ ⎭

Outer loop: time stepping

2. Calculations at each time instant

t = i∆t, i = 1, 2, 3, …
Initialise j = 1, {u}i(j) = {u}i–1
(j)
, {fs}i(j) = {fs}i–1
(j) (j) k
, ks = s_initial

Internal loop: iteration

4. Formulate the global stiffness

15. State determination matrix
K*  K*ab
15.1 {f*s}i(j+1) = {fs}i(j) + {Fs}i(j) 3. Spring stiffness 4.1 [K*] = [K] + [ks]i(j) = aa
K*ba K*bb
15.2 {u}i(j+1) = [{ua}i(j) {ub}]T j=j+1 (j) ks(j) Ktt Kt0
3.1 [ks]i(j) = ks – 4.2 = [K*aa]
(Fs)i(j) – (Fs)i–1 – ks(j) ks(j) K0t K00
15.3 {ks}i(j+1) =
ua ut

���
���

���
���
(us)i(j) – (us)i–1 4.3 = {u}i(j) = {ua}i(j)
ub u0
14. Force
Fs
5. Static condensation – update stiffness
Mtt0 üt C 0 u̇t K  K ut P

���
���

���
���

���
���

���
���
–us us 5.1 + tt + tt t0 = t
13. Footing
0 0 ü0 0 0 u̇0 K0t K00 u0 P0
displacement
CONSTANT UPDATE
us = {ua1}i(j)
fst K  K ut
���
���

���
���
5.2 = tt t0 = {f*s}
fso K0t K00 u0
~
5.3 {f s} = {fst} – [Kt0][K00]–1{fso}

6. Earthquake loading
6.1 {Pe } = –[Mtt]{ιh}üg(t)

11. Total displacements

12. Elastic state of
11.1 {ut}i(j) = {ut}i(j–1) + {dut}i(j) 7. Effective loading on structure
frame structure ~
11.2 {uo}i(j) = [Koo]–1{P0} – 7.1 {P }i = {Pe}i + {Pt}i – [Kto ][K00 ]–1{P0}
12.1 [K] = Kaa Kab
[Koo]–1[Kot]{ut}i(j) Kba Kbb 7.2
~
{P̂}i = {P }i + [a1]{ut}i–1 + [a2]{u̇t}i–1 + [a3]{üt}i–1
11.3 {du}i(j) = {dut}i(j) + {duo}
12.2 {fs}i(j) = [Kaa]{ua}i(j)
11.4 {ua}i(j) = [{ut}i(j), {uo}i(j)]T
8. Convergence check
~
8.2 {R̂}i(j) = {P̂i}i – {f s}i(j) – [a1]{ui}i(j)
8.3 ||R̂i(j)||∞ = max {| R̂i(j)|}
10. Solve 9. Newmark’s stiffness matrix
~
10.1 {dut}i(j) = [Kˆi(j)]–1 {R̂i(j)} 9.1 [K ]i(j) = [Ktt] – [Kto] [Koo]–1[Kot]
~
10.2 {duo} = –[Koo]–1[Kot]{dut}i(j) 9.2 [Kˆ ]i(j) = [K ]i(j) + [a1] ||R̂i(j)||∞ > Γ ||R̂i(j)||∞ ≤ Γ

17. Energy calculations and damage

({u}i – {u}i–1) + ⎫⎪1 – ⎫⎪{u̇}i–1 + ∆t ⎫⎪1 – ⎫⎪{ü}i–1

16. Velocity and acceleration
17.1 (EM)i , (EC)i , (EK)i

⎭ β⎭ ⎭ 2β⎭
γ γ γ
i∆t 1 16.1 {u̇}i =
17.2 (EH)i = ∑ i=1 ((Fs)i + (Fs)i–1)((ds)i + (ds)i – (ds)i–1)
({u̇}i–1 – ⎫⎪ – 1⎫⎪{ü}i–1
β∆t
2

⎭2β ⎭
γ 1 1
dm β 16.2 {ü}i = ({u} – {u}i–1) –
17.3 DI = + E with dm = max (|ds|) β(∆t)2 i β∆t
du v y du H i=0i∆t

Figure 7 Pseudo-dynamic numerical model algorithm

elastic stiffness of the footing and stub col- Reinforced concrete is a highly non- each time step, an initial out-of-balance
umn can also result in instability problems linear material, and with the application residual force vector was produced. As
in the convergence of the solution. of the applied loading at the beginning of a result, the structure was no longer in

54 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Table 1 Results from pseudo-dynamic analysis at 5% damping and a natural period of vibration of 0.86 s
Cracking (1) Yielding (1) Maximum
Intensity
PGA (g) Fc uc Fy uy Fm um Damage state
(MMI)
(kN) (mm) (kN) (mm) (kN) (mm)

7.45 0.34 22.1 1.68 – – 48 5.72 Onset of cracking, still serviceable

8.29 0.68 40.5 2.48 67.4 6.85 70.3 36.3 Large cracks, extensive damage

8.45 0.78 45.8 2.75 74.6 7.21 74.6 62.7 Collapse

8.75 1 40.84 3.37 62.0 7.41 74 62.7 Collapse

9.59 2 (1.21(2)) 35.8 2.47 71.5 7.81 71.6 62.2 Collapse

(1) Minimum lateral force and displacement that result in cracking of the concrete and yielding of the reinforcement
(2) The maximum acceleration achieved by the structure before failure

force or energy equilibrium. In general,
the displacement applied to the structure
results in a restoring force that differs
from that of the restoring force calculated
using the initially assumed stiffness of the
structure. Because of this, the stiffness of
the footing and stub column (ks) between
the previous time step and the current
time step needed to be updated at each
iteration until convergence was reached,
which was achieved once the infinity norm
was reduced to below the prescribed stop
criteria (r) shown in Figure 7. This needs
to be done to ensure force and energy
equilibrium at each time step. The HBM
data acquisition system has a 16-bit digital-
to-analogue converter (DAC) and produced
an output resolution of 0.003 mm for the
horizontal actuator with a stroke length
of 200 mm (shown previously in Figure 5)
that was used to apply the load onto the
specimen. Therefore, the stop criteria in
the research, indicated by (r) in Figure 7
was selected to be less than the resolution
of the data acquisition system. Once the
solution converged, and equilibrium within
the structure was achieved, the next time
step “i” could commence.
0.34 g to 2 g. Only minimal cracking had
occurred at a PGA of 0.34 g, with complete
failure having occurred at a PGA of 2 g.
Figures 18 and 19 show the damage to the
reinforced concrete stub columns during
the 0.78 g PGA test. The column continued
to carry the axial load until the cross-
sectional area had reduced substantially,
resulting in the axial capacity of the stub
columns being exceeded. The structure
subjected to the 0.78 g PGA test collapsed
due to the formation of a plastic hinge and
the loss of axial load carrying capacity.
Figure 18 shows the buckled and fractured
reinforcement due to excessive lateral
deformation of the stub column during the
0.78 g PGA test, and Figure 19 shows the
formation of the plastic hinge at collapse of
the stub column.
The results produced during the pseu-
do-dynamic experimentation were validat-
ed by replacing the physical test setup with
bilinear hysteretic shear models to observe
the expected damage over a range of PGAs.
During the experiments, convergence was
60
achieved for each time step for the duration
of the applied earthquake record, and the
damage observed to the footing during the
experiment corresponded with hysteretic
curves and the absorbed energy due to the
plastic deformation and buckling of the
reinforcement, concrete spalling and con-
crete crushing. The calibrated strain gauges
attached to the reinforcement indicated the
onset of reinforcement yielding that corre-
sponded to the flattening of the hysteretic
curve. However, the strain gauges failed
at larger deformations and were therefore
not able to compare the calculated energy
absorbed from the hysteretic curve with
the results from the strain gauges. The
displacements obtained from the external
LVDTs were used to ensure that the dis-
placement applied by the servo controller
was correct at each time step.
Table 1 summarises the results obtained
during the pseudo-dynamic experi-
ments for the following critical points:
the minimum cracking force (Fc) and
cracking deformation (uc), the minimum

40
Results of the pseudo-
dynamic experiments 20
The hysteretic behaviour of each of the
Load (kN)

reinforced concrete stub columns is provid-

0
ed and the observed damage is discussed
in terms of cracking, concrete spalling,
reinforcement buckling, and fracturing for –20
each of the pseudo-dynamic experiments.
–40
Hysteretic relations and
crack patterns
–60
The hysteretic results and force-versus- –10 –8 –6 –4 –2 0 2 4 6 8 10
time results for each of the experiments Displacement (mm)
are shown in Figures 8 to 17, with the peak
ground accelerations (PGA) ranging from Figure 8 Hysteretic response under El Centro earthquake scaled to 0.34 g

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 55
 50

40

30

20

10
Force (kN)

–10

–20

–30

–40

–50
0 5 10 15 20 25 30 35 40 45 50 55
Time (s)

Figure 9 Force-versus-time graph for the 0.34 g experiment

reinforcement yielding force (Fy) and defor-

mation (uy), and the maximum achieved 100
force (Fm) and deformation (um). The fol- 80
lowing observations were made during the
60
pseudo-dynamic experiments:
40
QQ With increasing deformation, the shear
Load (kN)

load-carrying capacity tends to a constant 20

value with a subsequent reduction in the 0
lateral stiffness of the footing and stub col-
–20
umn, and therefore there is a maximum
–40
shear load that the structure is capable of
transferring between the ground and the –60
structure as deformation increases. –80
QQ The lateral capacity of the reinforced
–100
concrete stub column and subsequent –40 –30 –20 –10 0 10 20 30 40
damage are predominately controlled Displacement (mm)
by the yield strength and ductility of the
reinforcement. Figure 10 Hysteretic response under El Centro earthquake scaled to 0.68 g

80

60

40

20
Force (kN)

–20

–40

–60

–80
0 5 10 15 20 25 30 35 40 45 50 55
Time (s)

Figure 11 Force-versus-time graph for the 0.68 g experiment

56 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 QQ An increase in the applied lateral
100 deformation applied to the stub column
80 results in cracking and yielding of the
reinforcement, which in turn result in a
60
reduction in the lateral stiffness of the
40 stub column.
20
QQ The ductility of the reinforced con-
Load (kN)

crete stub column is controlled by the

0
tensile reinforcement and the shear
–20 reinforcement.
QQ Cracks occurred near the base of the
–40
column where the maximum moment
–60 was expected; however, the cracks
–80 did not always open at the interface
between the base of the column and the
–100
–80 –60 –40 –20 0 20 40 60 80 top of the stub column.
Displacement (mm) QQ The number of cycles of vibration
increases the damage to the reinforced
Figure 12 Hysteretic response under El Centro earthquake scaled to 0.78 g concrete stub column, particularly when
the load reverses after the reinforcement
has yielded.
80 QQ Spalling of concrete occurs due to
the buckling of reinforcement during
60 load reversal from tensile loading to
compression loading on either face of
40
the concrete column in the direction
20
of loading, thus resulting in a reduc-
tion in shear capacity with each cycle
Force (kN)

0 as indicated in Figure 12. This occurs

due to the incompatibility between the
–20 brittle concrete material and the ductile
reinforcement. Upon load reversal,
–40 the permanently elongated reinforce-
ment is first mobilised in compression
–60
before the crack that has formed in
the concrete can close and mobilise in
–80
0 1 2 3 4 5 6 compression. To overcome this incom-
Time (s) patibility, the reinforcement buckles,
resulting in the spalling of the concrete.
Figure 13 Force-versus-time graph for the 0.78 g experiment The spalling of the concrete concomi-
tantly results in a reduction in the gross
cross-sectional area of the column,
100 which in turn results in a decrease in
80 the axial and shear capacity of the stub
column.
60
QQ At high peak ground accelerations, the
40 failure tends to that obtained during
20 quasistatic tests with little to no con-
Load (kN)

crete spalling and the failure governed

0
by fracturing of the reinforcement and
–20 crushing of the concrete as shown in
Figures 14 and 16. This results in little
–40
to no reduction in shear capacity prior
–60 to failure.
–80 QQ The unloading stiffness from the back-
bone curve (the curve produced during
–100
–60 –50 –40 –30 –20 –10 0 10 20 30 40 50 60 quasistatic experiments) of the stub
Displacement (mm) column is greater than the reloading
stiffness into the backbone curve, which
Figure 14 Hysteretic response under El Centro earthquake scaled to 1 g

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 57
 indicates that the structure is absorbing
energy and incurring damage. 80
QQ Before the reinforcement yields, the
response under cyclic loading remains 60

predominantly perfectly plastic without

40
any significant permanent deformation.
QQ The loss of moment capacity due to the
20
formation of a plastic hinge is governed

Force (kN)
by the repeated cycling of the stub 0
column at a displacement less than the
quasistatic loading failure displacement –20
and greater than the yielding displace-
ment of the reinforcement. –40
QQ Increasing the amplitude of the El
–60
Centro ground motion record showed
that there is a maximum PGA that can
–80
be sustained by the stub column before 0 0.5 1.0 1.5 2.0 2.5 3.0
failure occurs. For example, the pseudo- Time (s)
dynamic test that was undertaken by
amplifying the El Centro ground motion Figure 15 Force-versus-time graph for the 1 g experiment
record to a PGA of 2 g only managed
to achieve a maximum PGA of 1.21 g
before failure occurred. 100

80
Energy-related results
60
The amount of energy imparted to the
structure due to the earthquake is dis- 40
tributed between the kinetic energy (E M), 20
Load (kN)

damping energy (Ec), strain energy (EK)

0
and hysteretic energy (EH). This section
shows the distribution of energy within the –20
structure for each of the pseudo-dynamic
–40
tests. Figure 20 shows the time history of
the total energy imparted to the structure –60

for the duration of each of the scaled –80

earthquake ground motion records during
–100
the pseudo-dynamic experiments and is –70 –60 –50 –40 –30 –20 –10 0 10 20 30 40 50 60 70
either shown for the entire duration of Displacement (mm)
the earthquake record or until structural
failure of the reinforced concrete stub col- Figure 16 Hysteretic response under El Centro earthquake scaled to 2 g
umn. The experiments conducted at 0.34 g
and 0.68 g ran for the full duration of the
earthquake record, whereas the 0.78 g, 80
1 g and 2 g experiments all failed before
60
completion of the earthquake record.
The maximum energy imparted to the 40
structure occurred during the experiment
with a maximum peak ground accelera- 20
Force (kN)

tion of 0.68 g, and the tests undertaken at

0
larger peak ground accelerations showed
a reduction in the total energy imparted
–20
to the structure before failure of the stub
column occurred. –40
Figure 21 shows the total hysteretic
energy absorbed as time progressed with the –60
application of the scaled El Centro ground
–80
motion record for each of the pseudo- 0 0.5 1.0 1.5 2.0 2.5
dynamic tests. Table 2 shows the distribu- Time (s)
tion of energy within the structure for each
of the amplified ground motion records, Figure 17 Force-versus-time for the 2 g experiment

58 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 (a) (b)

Figure 18 Resultant damage to the reinforced concrete stub column with (a) outward buckling of the reinforcement and (b) reinforcement fracturing

(a) (b)

Figure 19 Plastic hinge formation at the base of the column at collapse

which is either recorded at the end of the

300 earthquake record or at failure of the stub
column. Table 2 also provides the amount
of energy absorbed by the stub column as
250
a percentage of the total energy imparted
to the frame structure. The stub columns
that showed more substantial observed
200
damage during the experiments absorbed
the largest quantity of energy. The 0.78 g
Energy (kJ)

150 ground motion record resulted in the most

observed damage and absorbed the greatest
quantity of energy as is seen in Figure 21.
100 Even though the 1 g and 2 g ground motion
records resulted in the failing of the struc-
ture, the amount of energy absorbed by the
50 stub column is less than that absorbed by
the 0.68 g and 0.78 g earthquake. Therefore,
the hysteretic energy absorbed by the stub
0
0 10 20 30 40 50 60 column gives a good indication of damage
Time (seconds) incurred by the stub column under repeated
PGA 0.34 g 0.68 g 0.78 g 1g 2g cyclic loading, but does not indicate struc-
tural failure at large deformations with few
Figure 20 Total energy imparted to the structure at the scaled peak ground accelerations or no cycles of vibration.

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 59
 Figures 22 to 26 show the energy
components for the frame structure for 14
the entire duration of the ground motion
record that were scaled to produce a 0.34 g 12
and 0.68 g peak ground acceleration. The
structure subjected to the 0.34 g peak
10
ground acceleration resulted in damping
absorbing all the energy imparted to the

Energy (kJ)
structure. The 0.68 g peak ground accelera- 8
tion resulted in the stub column absorbing
hysteretic energy and therefore indicates a 6
correlation between the number of cycles of
vibration and the resultant damage in terms
of observed concrete spalling, yielding and 4

buckling of the reinforcement. The ductility

provided by the reinforcement has a signifi- 2
cant influence on the quantity of energy that
can be absorbed by the stub column before
0
the failure displacement is exceeded. 0 5 10 15 20 25 30 35 40 45 50 55 60
The test at 0.34 g did not absorb a large Time (seconds)
amount of energy as the stub column did PGA 0.34 g 0.68 g 0.78 g 1g 2g
not undergo significant lateral deformation
and the reinforcement did not yield. For the Figure 21 Hysteretic energy of the reinforced concrete stub columns
tests at peak ground accelerations of 0.68 g
and 0.78 g, the stub columns underwent a Table 2 Distribution of energy at the end of the ground motion record or at failure
larger number of cycles of vibration in the Total Percentage
plastic region of the reinforcement, which Hysteretic Inertia Damping Strain
PGA energy energy
(EH) (EM) (EC) (EK)
consequently resulted in more damage to (g) imparted absorbed
(kJ) (kJ) (kJ) (kJ)
(kJ) (%)
the stub column due to the spalling of the
concrete cover. The spalling of the concrete 0.34 77.3 0.02 0.0 77.0 0.3 0.03
cover occurred predominately due to the 0.68 298.8 9.1 0.0 289.4 0.3 3.1
buckling of the permanently elongated
0.78 259.9 13.2 5.3 182.4 59.0 5.1
reinforcement upon load reversal from ten-
sion to compression, and to a lesser extent 1 169.6 4.2 1.1 56.3 108.0 2.5
due to compression capacity of the concrete 2 (1.21)* 158.6 3.1 0.2 33.0 122.3 2.0
being exceeded. The tests carried out at 1 g
* The maximum acceleration achieved by the structure before failure
and 2 g had fewer or no cycles of vibration
before the reinforcement fractured and visu-
ally did not experience the same amount of
damage in terms of concrete crushing and 90
spalling of the concrete cover than what was
80
observed from the tests conducted at lower
peak ground accelerations. 70
It is evident that most of the energy
absorbed in the structure is due to Rayleigh 60
damping, with only a small percentage
Energy (kJ)

of the overall energy absorbed by the 50

reinforced concrete stub column. The
40
hysteretic energy correlates well with the
cumulative damage to the reinforced con- 30
crete stub column and the observed dam-
age experienced by the specimens during 20
the pseudo-dynamic experiments.
10

0
Conclusions 0 5 10 15 20 25 30 35 40 45 50 55 60
The pseudo-dynamic experimental method Time (seconds)
using Newmark’s implicit time-stepping EH EH + EC EH + EC + EK EH + EC + EK + EM
method was presented. It is concluded that
the method provides a viable approach to Figure 22 Time histories for energy terms under the El Centro earthquake scaled to 0.34 g

60 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
 greater than the maximum achievable stiff-
300 ness of the specimen being tested to ensure
the stability of the analysis and prevent the
formation of premature damage.
250
By using the pseudo-dynamic method,
the hysteretic response of the footing and
200 stub columns could be related to earthquake
intensity, and observations could be made on
Energy (kJ)

the extent of damage incurred by the stub

150
column due to the applied earthquake load-
ing. The reinforced concrete stub column
100 only dissipates a small percentage of the over-
all energy imparted to the structure in the
form of hysteretic energy, with the remainder
50 of the energy being dissipated due to damp-
ing within the frame structure, which con-
sists of a number of structural elements. The
0
0 5 10 15 20 25 30 35 40 45 50 55 60 extent of damage to the reinforced concrete
Time (seconds) stub column is governed predominately by
EH EH + EC EH + EC + EK EH + EC + EK + EM the yield strength of the reinforcement, with
the rate of damage and hysteretic energy
Figure 23 Time histories for energy terms under the El Centro earthquake scaled to 0.68 g absorption increasing substantially once
the reinforcement has yielded. The research
showed that lower PGAs tend to result in
300 the absorbed hysteretic energy and cyclic
behaviour of the stub column governing
the failure of the stub column, whereas at
250
large PGAs the fracturing of the reinforce-
ment tends to dominate the failure of the
200 structure. Therefore, using pseudo-dynamic
experimentation on a number of structural
Energy (kJ)

members can be used to determine the risk

150
of damage and failure of a structural member
with increasing earthquake intensity.
100

50
0
0 1 2 3 4 5
Time (seconds)
EH EH + EC EH + EC + EK EH + EC + EK + EM
Figure 24 Time histories for energy terms under the El Centro earthquake scaled to 0.78 g
correlate damage to a physical test model of results and strain energy absorbed during
a reinforced concrete footing and stub col- the applied earthquake. The use of the
umn, which forms part of an overall struc- implicit Newmark’s method provided a sta-
ture, with increasing earthquake intensity. ble and accurate algorithm to quantify the
The hysteretic curves produced from the damage incurred to the reinforced concrete
pseudo-dynamic experiment were validated stub column due to the overall response
through the combination of strain gauges of the structure that had been subjected
attached to the reinforcement, LVDTS and to an applied seismic load. Convergence to
running the pseudo-dynamic algorithm a solution was achieved at each time step
using bilinear hysteretic numerical shear for the duration of the applied earthquake,
model to ensure that convergence to a solu- thus ensuring energy and force equilib-
tion was achieved. The damage observed rium, further validating the method. The
during the pseudo-dynamic experiments initial stiffness that is used within the
also corresponded with the hysteretic implicit Newmark’s algorithm must be
Recommendations for
further work
Cognisance should be given to the fact that
the analysis undertaken during this research
6 only considered a single reinforced concrete
stub column design that was analysed using
a single earthquake ground motion record.
Therefore, recommendation for future work
would include the following:
QQ Analysing the structure by varying the
longitudinal and shear reinforcement
to determine the influence it has on the
damage incurred by the stub column.
QQ Testing reinforced concrete stub col-
umns with closed seismic stirrups and
comparing the capacity with the results
produced using traditional stirrups.
QQ Performing pseudo-dynamic experi-
ments that account for both a varying
shear and axial loading as could be
experienced by foundations placed on
the exterior of a building.
QQ Determining the influence that com-
pacted soil around the stub column

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 61
 has on the response of the reinforced
concrete stub column and the resultant 180
damage, therefore incorporating soil-
160
structure interaction into the analysis.
QQ Performing pseudo-dynamic experi- 140
ments that enable the analysis of the
reinforced concrete stub column under 120

biaxial bending and shear.

Energy (kJ)
100
QQ Performing pseudo-dynamic tests on
reinforced concrete stub columns by 80
accounting for non-linear behaviour
60
within the overall structure and deter-
mining what influence the softening of 40
the overall structure will have on the
performance of the stub column. 20

0
0 0.5 1.0 1.5 2.0 2.5 3.0
Acknowledgements Time (seconds)
This research has been funded by MMI EH EH + EC EH + EC + EK EH + EC + EK + EM
Holdings Limited and the University
of Pretoria’s Natural Hazard Centre, Figure 25 Time histories for energy terms under the El Centro earthquake scaled to 1 g
NRF THRIP Funding TP 14072278140
and NRF funding Grant: 103724 and
Grant: TP14072278140. These sponsors are 180
gratefully acknowledged for their contribu-
160
tions. Any opinions, conclusions and recom-
mendations are those of the authors and do 140
not necessarily reflect those of the sponsors.
120
Energy (kJ)

References
Brandt, M 2011. Seismic hazard in South Africa. Report
No 2011–0061. Pretoria: Council for Geoscience.
Chang, S 2009. Bidirectional pseudodynamic
testing. Journal of Engineering Mechanics, 135(11):
1227‒1236.
Chopra, A K 2012. Dynamics of structures. Theory
and Applications to Earthquake Engineering. Upper
Saddle River, NJ: Prentice Hall.
HBM (Hottinger Baldwin & Messtechnik GmbH) 2016.
Catman Data Acquisition Software – CatmanAP,
Version V4.2.2.14. Darmstadt, Germany: HBM.
Kurt, E G 2010. Investigation of strengthening
techniques using pseudo-dynamic testing. Ankara,
Turkey: Middle East Technical University.
Mosalam, K M, White, R N & Ayala, G 1998.
Response of infilled frames using pseudo-dynamic
experimentation. Earthquake Engineering and
Structural Dynamics, 27: 589‒608.
Mosalam, K M, White, R N & Gergely, P 1997. Seismic
evaluation of frames with infill walls using pseudo-
dynamic experiments. Technical Report NCEER-97-
0020. Ithaca, NY: Cornell University.
Pegan, P & Pinto, A V 2000. Pseudo-dynamic testing with
substructuring at the ELSA Laboratory. Earthquake
Engineering and Structural Dynamics, 29: 905‒925.
SANS (South African National Standard) 2000. SANS
10100-1. The Structural Use of Concrete. Part 1.
Design. Pretoria: SABS Standards Division.
100
80
60
40
20
0
0 0.5
EH EH + EC
Figure 26 Time histories for energy terms under the El Centro earthquake scaled to 2 g
SANS (South African National Standard) 2017. SANS
10160-4. Basis of Structural Design and Actions for
Buildings and Industrial Structures. Part 4. Seismic
Actions and General Requirements for Buildings.
Pretoria: SABS Standards Division.
Shing, P B & Mahin, S A 1984. Pseudodynamic test
method for seismic performance evaluation: Theory
and Implementation. Report No UCB/EERC-84/01.
University of California, Berkeley: Earthquake
Engineering Research Centre.
Takanashi, K, Udagawa, K, Seki, M, Okada, T &
Tanaka, H 1975. Non-linear earthquake response
analysis of structures by a computer-actuator
on-line system. University of California, Berkeley:
Earthquake Engineering Research Centre.
1.0 1.5 2.0 2.5
Time (seconds)
EH + EC + EK EH + EC + EK + EM
Takanashi, K & Nakashima, M 1987. Japanese
activities on on-line testing. Journal of Engineering
Mechanics, 113(7): 1014‒1032.
Thewalt, C R & Mahin, S A 1987. Hybrid solution
techniques for generalised pseudodynamic testing.
Report No UCB/EERC-87/09. University of
California, Berkeley: Earthquake Engineering
Research Centre.
Wang, T, Nakashima, M & Pan, P 2006. On-line hybrid
test combining with general-purpose finite element
software. Earthquake Engineering and Structural
Dynamics, 35: 1471‒1488.

62 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
What leads to severe TECHNICAL PAPER
multi-vehicle crashes on Journal of the South African
Institution of Civil Engineering

mountainous expressways ISSN 1021-2019

Vol 64 No 1, March 2022, Pages 63–70, Paper 1174

in Western China? PROF YONGGANG WANG (Pr Eng) is a full

professor of Transportation Engineering at
Chang’an University, and currently serves as
an editorial board member of Proceedings of
Y Wang, L Wang, L Sun the Institution of Civil Engineers – Transport,
Scientia Iranica, Transport and Frontiers in
Public Health. He obtained his BSc and
Master’s in Civil Engineering from
This paper investigates the occurrence and severity of collisions involving multiple vehicles on Shijiazhuang Tiedao University (2001) and Ocean University of China
mountain expressways (MMEs) in Western China. A total of 1 521 crash samples occurring on (2004), respectively, and his PhD in Transportation Engineering from
one typical mountain expressway in Shaanxi, China, between 2012 and 2017, were analysed Harbin Institute of Technology (2009). His research interests cover traffic
crash data modelling, driving behaviour analysis, and related topics. He
through a partially constrained generalised ordered logit to identify the significant risk factors
has published more than 40 papers in international journals (ORCID:
contributing to the severity of such crashes. Elasticity analysis was performed to quantify the 0000-0002-9365-1851).
effects of each independent explanatory variable on the collision severity outcomes. Fourteen
Contact details:
total explanatory variables were found to have a significant and pronounced influence on the College of Transportation Engineering
likelihood of MME crashes. These include the type of collision, the at-fault driver’s age, driving Chang’an University
while fatigued, cell phone use while driving, alcohol-impaired driving, speeding, risky following South 2 Ring Road, Xi’an 710064, China
T: +86 29 8233 4050
and dangerous overtaking behaviour, sharp curves in the roadway and slippery pavement E: wangyg@chd.edu.cn
conditions, seasons, day of the week, time of day, and adverse weather (rain/snow/fog). The
impacts of the variables on the collision severity were also explored. Taken together, the LONGJIAN WANG obtained his BSc in
findings may serve as a useful guide for developing legislation and technical countermeasures Electronic Information Engineering from
to ensure traffic safety on mountain expressways in Western China. Heilongjiang University (2013) and his
Master’s in Transportation Engineering from
Chang’an University (2017), where he is
currently a PhD student in Transportation
INTRODUCTION on mountainous expressways, due to the Engineering. His research interests include
The number of registered motor vehicles adverse traffic environment of the terrain, transportation planning, traffic safety and
driving behaviour. He has published several papers in international
has increased dramatically in China over which includes tight curves, steep slopes, journals and book series.
the past two decades. Numbers have the existence of bridges and tunnels, and
Contact details:
soared from about 9.6 million in 2003 to changing climatic conditions (Meng 2017; College of Transportation Engineering
more than 327 million in 2018, an almost Wang & Prato 2019; Wang et al 2019a). Chang’an University
32-fold increase (National Bureau of The huge economic loss and serious social South 2 Ring Road, Xi’an 710064, China
T: +86 29 8233 4067
Statistics of China 2019), which in turn repercussions that are incurred by crashes
E: wanglj@chd.edu.cn
resulted in a great number of road motor involving multiple vehicles have attracted
traffic crashes (Benlagha & Charfeddine increasing attention worldwide from LETIAN SUN is a final-year undergraduate
2020). In 2018 a total of 244 937 road researchers and traffic managers (Wang student in transportation engineering at
motor traffic crashes occurred, with & Prato 2019; Wang et al 2019a; Dong et Chang’an University. His research interests
focus on traffic safety, traffic flow theory,
258 532 injuries, 63 194 fatalities, and al 2018; Rezapour et al 2019), especially
and connected and autonomous driving
direct economic loss of 0.221 billion US for those occurring on mountainous techniques. He has won prizes in the
dollars in China. A large proportion of expressways. Therefore, the potential National Competition of Transport Science
these records were identified as occur- risk factors associated with these crashes and Technology for Undergraduate
Students, and in the National Undergraduate Training Program for
ring along roads in mountainous areas must be identified to better understand
Innovation and Entrepreneurship.
(National Bureau of Statistics of China how they occur, and to suggest suitable
Contact details:
2019). Alarming statistics in China show countermeasures.
College of Transportation Engineering
that mountainous expressways are suscep- In recent years, considerable research Chang’an University
tible to a high frequency of multi-vehicle efforts have focused on investigating the South 2 Ring Road, Xi’an 710064, China
crashes, as well as more severe conse- geometric characteristics of roadways that T: +86 29 8233 4066
E: sunltian@163.com
quences (Zhang et al 2016; Meng 2017). may contribute to the occurrence of MME
Unlike single-vehicle crashes resulting crashes (Yu et al 2015). For example, Rusli
from loss of vehicle control associated et al (2018) examined MME crashes in
with driver error or negligence like exces- Malaysia and found that the presence of
sive speed, alcohol usage and driving minor junctions enhances the likelihood
fatigue (Rusli et al 2017), it is extremely of MME crashes, while the existence of
Keywords: multiple-vehicle crash, mountain expressway, risk factor,
difficult to determine the causes of multi- horizontal curves along a steep gradient, partially constrained generalised ordered logit model,
vehicle crashes, especially when occurring and the presence of a passing lane, increase elasticity analysis

Wang Y, Wang L, Sun L. What leads to severe multi-vehicle crashes on mountainous expressways in Western China?
J. S. Afr. Inst. Civ. Eng. 2022:64(1), Art. #1174, 8 pages. http://dx.doi.org/10.17159/2309-8775/2022/v64no1a6 63
the likelihood of such crashes (Rusli et al
2018). Additionally, weather and traffic-
related factors, such as rainfall (Rusli et al
2018), visibility (Yu et al 2015), pavement
surface conditions (Ma et al 2016; Wang et
al 2021b) and annual average daily traffic
(Sameen & Pradhan 2017), influence the
occurrence of MME crashes significantly.
In Chongqing, China, the season, time of XJ IM
the crash, involvement of trucks, crash GS NX
features, weather, and roadway conditions
have been found to have obvious impacts QH
on the levels of MME crashes (Meng SX
2017). Driver factors are also frequently TI
explored by researchers. Zhang et al CQ
SC
(2016) investigated MME crash data from
the Taigan Expressway in Jiangxi, China,
finding that younger and older drivers,
GZ
especially female drivers, contribute more YN GX
to severe MME crashes (Meng 2017). Rusli
et al (2018) showed that driver speeding
significantly increased the injury sever-
ity in MME crashes. Wang et al (2019a)
and Dong et al (2018) found that truck Figure 1 The western region of China mainland
drivers’ age, seatbelt status, and speeding
and risky following behaviour signifi-
cantly correlated with severe crashes on
mountainous expressways.
According to China’s administrative
division, China’s western region embraces
the six provinces of Shaanxi (SX), Gansu
(GS), Qinghai (QH), Sichuan (SC), Yunnan
(YN) and Guizhou (GZ); the five autono-
mous regions of Guangxi (GX), Ningxia
(NX), Inner Mongolia (IM), Xinjiang (XJ)
and Tibet (TI); and the Chongqing (CQ)
Municipality as the red-marked area in
Figure 1 (excluding the South China Sea) ‒
a land area of 6.86 million square kilome-
tres, accounting for 70.6% of the country’s
total land area. Much of the Western
China territory (81.9%) is mountainous,
and unfortunately few studies over the past
several years have focused on a quantitative
impact analysis of the various potential Figure 2 One typical mountainous expressway segment in Shaanxi Province, China
factors that contribute to MME crashes
in this vast area (Wang et al 2021a) where China, over a recent six-year time frame, METHODOLOGY
at-fault driver behaviour, geometric design the primary purpose of this study was
elements, and environmental conditions therefore to apply a generalised ordered Data collection
are unique from other areas. logit model to quantify the potential risk A total of 1 705 police-reported multi-
Shaanxi Province (SX) ‒ located in factors associated with the severity of vehicle crashes between 2012 and 2017,
the heart of China, straddling the Loess MME crashes, the severity of collisions accounting for 63.01% of the total obser-
Plateau, Guanzhong Plains and Qinba resulting from these crashes, and the mar- vations, were collected from the Xihan
Mountains from north to south, with ginal effects of each explanatory factor on Expressway in Shaanxi (SX), China.
complex topography and landscape ‒ has MME crash severity. It is anticipated that This stretch of road (shown in Figure 2)
seen a representative number of MME the findings presented here can be used is a four-lane 198.5 km segment of the
crashes (Wang & Prato 2019; Wang et to guide the development of legislations G5 Jingkun Expressway from Laoyukou
al 2019a; 2021b). Using reported multi- and technical countermeasures for traffic Toll Station (K1131+657) to Xiejiaying
vehicle crash data from one typical safety on mountainous expressways in Interchange (K1330+204), with a speed
mountainous expressway in Shaanxi, Western China. limit of 60 ~ 80 km/h. Among the MME

64 Volume 64 Number 1 March 2022 Journal of the South African Institution of Civil Engineering
Table 1 Contributory variable description
Variable Code Frequency % Code Frequency %

Crash feature

1 = PDO 767 50.43

Severity 2 = Injury 463 30.44
3 = Fatality 291 19.13

1 = Head-on 173 11.37 2 = Rear-end 1040 68.38

Type of collision
3 = Sideswipe* 200 13.15 4 = Angle 108 7.10

Driver factor

Gender 0 = Female 141 9.27 1 = Male 1380 90.73

1 = Young* 355 23.34

Age 2 = Adult 1094 71.93
3 = Old 72 4.73

Driving while fatigued 0 = No 1403 92.24 1 = Yes 118 7.76

Cell phone use 0 = No 1441 94.74 1 = Yes 80 5.26

Alcohol use 0 = No 1438 94.54 1 = Yes 83 5.46

Vehicle factor

Truck involved 0 = No 991 65.15 1 = Yes 530 34.85

Speeding 0 = No 1383 90.93 1 = Yes 138 9.07

Risky following 0 = No 1345 88.43 1 = Yes 176 11.57

6.90
Dangerous overtaking 0 = No 1416 93.10 1 = Yes 105

Roadway factor

Sharp curve 0 = No 815 53.58 1 = Yes 706 46.42

Steep slope 0 = No 759 49.90 1 = Yes 762 50.10

Pavement condition 0 = Dry 1023 67.26 1 = Slippery 498 32.74

Environmental factors

1 = Spring* 298 19.59 2 = Summer 374 24.59

Seasons
3 = Autumn 472 31.03 4 = Winter 377 24.79

Day of week 0 = Working days 1032 67.85 1 = Weekends / holidays 489 32.15

1 = 06:00–18:00* 823 54.11

Time of day 2 = 18:00–24:00 344 22.62
3 = 24:00–06:00 354 23.27

Weather 0 = Clear 1223 80.41 1 = Adverse 298 19.59

*base

crash records, 184 cases were deleted due
to containing incomplete information,
meaning that 1 521 cases were included in
the final database.
A three-point ordinal scale was used to
classify the MME crash severities measured
by the most severely injured person(s),
including: (a) property damages only (PDO),
in which there were only damage to road
facilities and vehicles or negligible personal
injuries, (b) injury, in which there were
personal injuries requiring hospitalisation,
together with serious property damage,
and (c) fatality, in which there were persons
killed immediately or persons who died
within 30 days as a result of the crash
(Wang & Prato 2019; Wang et al 2019a).
The distribution of the crash severity
levels was as follows: PDO (50.43%), injury
(30.44%) and fatality (19.13%). Four types
of collisions were considered for further
analysis, namely, head-on, rear-end, side-
swipe, and angle.
Additionally, the crash database collected
information related to at-fault driver demo-
graphic characteristics and driving behaviour,
vehicle attributes, roadway conditions, and
environmental influence, as shown in Table 1.
The information correlated to the roadway
geometric factors was determined by the
original expressway design documents and
updated through the latest Google Earth,
and the rest was directly extracted from the
original accident reports released by the local
traffic management departments. Additional
information included:
QQ At-fault driver factors: gender, age (e.g.
< 30 years old, 30‒50 years old, and > 50
years old), driving fatigue, impairment
by alcohol, and cell phone usage
QQ Vehicle factors: truck involvement,
speeding and risky following, and dan-
gerous overtaking
QQ Roadway factors: sharp curves (radius of
horizontal curve < 2 000 m), steep slope
(longitudinal gradient > 3%), and slip-
pery pavement due to weather.
QQ Environmental factors: seasons (e.g.
spring: March to May, summer: June
to August, autumn: September to
November, and winter: December to
February), day of the week (including
working days = 0:00 Monday to 16:59
Friday, weekends/holidays = 17:00 Friday
to 24:00 Sunday, and public holidays
including New Year, Chinese New Year,
Qingming Festival, International Labour

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 65
 Day, Duanwu Festival, Mid-Autumn
Festival, and National Day), time of day
(e,g, 6:00‒18:00, 18:00‒24:00, 24:00‒6:00),
weather (including clear: sunny/cloudy,
and adverse: rainy/snowy/foggy).
Analytical model
Since the crash severity data is typically
ordinal, varying from a non-fatal to fatal
level in nature, traditionally ordered (both
probit and logit) probability models were
employed in the literature to model the
severity of traffic crashes (Zhang et al 2016;
Dong et al 2018; Rezapour et al 2019; Wang
et al 2021b; Kahn & Vachal 2020).
Let yi be the MME crash severity with
three categories (PDO, injury and fatal),
and xi be the potential variables affecting
the MME crash severity. A latent variable
yi* can then be used to measure the MME
crash severity through an ordered logit
approach:
yi * = x’i β + εi(1)
where xi = {1, xi1, xi2, …, xiN}T is a vector
representing the values of crash i on the
full set of N independent explanatory vari-
ables, β = {β 0, β1, β2, …, βN}T is a vector of
regression parameters to be estimated, and
εi is a random error term with standard
logistic distribution.
The relationship between the observed
levels of the dependent injury severity
yi and the latent injury risk yi* can be
expressed by introducing the thresholds α1
and α 2 as follows:
yi = 1 (PDO), if yi* << α1(2a)
yi = 2 (injury), if α1 < yi* << α 2(2b)
yi = 3 (fatal), if yi* > α 2(2c)
Thus, the probability P of MME crash
i having a severity level j can also be
expressed as:
P(yi > j) = g(x’i βj)
exp(αj – x’i βj)
= , j = 1,2,3
1 + exp(αj – x’i βj)
where αj is a cut-off point for the jth cumu-
lative logit.
It should be noted that Equation 1
should meet the parallel-lines assump-
tion, which requires that the estimated
parameters remain the same for different
severity levels (Wang et al 2019a). However,
such an assumption is often violated, so a
66
partially constrained generalised ordered
logit (PCGOL) model, also known as the
gamma parameterisation of partial propor-
tional odds model with logit function, was
proposed, which allowed the parallel-lines
assumption to be relaxed for one or a
few dependent variables but retained the
ordered nature for the majority of depen-
dent variables on a set of n independent
explanatory variables (Peterson & Harrel
1990) as:
P(yi > j) = g(x’iβj)
exp[αj –(x’i βj + z’i γj)]
= , j = 1,2,3(4)
1 + exp[αj –(x’i βj + z’i γj)]
where βj is a vector of coefficients correlat-
ed with a subset xi of independent explana-
tory variables (see Table 1) for which the
parallel-lines assumption is not violated,
and γj is a vector of coefficients correlated
with a subset zi of independent explana-
tory variables for which the parallel-lines
assumption is violated.
The violation of the parallel-lines
assumption was firstly checked for each
independent variable and then the two
parameter vectors βj and γj and cut-off
thresholds αj were estimated via the
maximum of the log likelihood function LL
(Peterson & Harrel 1990). In the proposed
model, each explanatory variable has one β
coefficient and (k-2)γ coefficients, where k
is 3 in the current research as the number
of alternatives. There were (k-1)α coef-
ficients reflecting the cut-off points.
Equivalently, Equation 4 can be rewrit-
ten using the cumulative probability
distribution as:
P(yi << j) =
 1 – g(x’i βj)
= F(αj – x’i βj), j = 1,2,3 (5)
which can also be expressed in Equations 6a
‒ 6c:
P(yi = 1) = F(α1 – x’i β1)(6a)
P(yi = 2) = F(α 2 – x’i β2) – F(α1 – x’i β1)(6b)
(3)
P(yi = 3) = 1 – F(α 2 – x’i β2)(6c)
Elasticity analysis
Additionally, each independent variable
(see Table 2) is transferred into a binary
categorical explanatory variable in deter-
mining the partial proportional odds
model; the elasticity cannot be measured
since it is not differentiable, so the direct
Volume 64 Number 1 March 2022
pseudo-elasticity analysis is conducted to
quantify the marginal effect of independent
variable n on the probability of severity
level j for MME crash i. The percentage
change in probability specific to severity
level j for MME crash i was calculated
when the nth binary variable xjin (n << N)
was switched from 0 to 1 or vice versa
(Wang & Prato 2019):
P(yi > j)[given xjin = 1] –
P(yi > j)[given xjin = 0]
ExP(y
jin
i >j) = (7)
P(yi > j)[given xjin = 0]
The pseudo-elasticities were calculated for
each severity level j and MME crash i, and
consequently averaged for each MME crash
severity j over all crash samples.
RESULTS AND DISCUSSION
Model estimation
The partial proportional odds model
was estimated via a user-written gologit2
procedure in Stata 15 statistical software
(Peterson & Harrel 1990), when explana-
tory variables were progressively added to
the model while testing the violation of the
parallel-lines assumption using the 0.05
level of significance. Such an interactive
procedure was performed to find the best
model until no further variable signifi-
cantly improved the fit of model. Finally,
the best fit model is presented in Table 2.
Fourteen total explanatory variables,
including type of collision, at-fault driver’s
age, driving while fatigued, cell phone use
while driving, alcohol-impaired driving,
speeding, risky following and dangerous
overtaking behaviour, sharp curves in the
roadway and slippery pavement condi-
tions, seasons, day of the week, time of day
and adverse weather, were all found to be
significantly associated with MME crash
severity. Four variables, namely at-fault
driver’s speeding, overtaking behaviour,
sharp curves, and time of day violated the
proportional odds assumption (see Table 2).
The marginal effects of each explanatory
variable on the probability of MME crash
severity level at 95% confidence level are
presented in Table 3.
Collision characteristics
The type of collision was classified into
four categories: head-on, rear-end, side-
swipe and angle, with sideswipe as the ref-
erence category. Significant difference was
observed between head-on and sideswipe
Journal of the South African Institution of Civil Engineering
Table 2 Estimation results of the PCGOL model Driver factors
Explanatory variables Est. S.E. Explanatory variables Est. S.E. The at-fault driver’s age was divided into
three levels ‒ less than 30 years, 30‒50
Beta Time of day 18:00-24:00 0.733** 0.229
years and more than 50 years, and the first
Type of collision Head-on 1.259** 0.401 24:00-06:00 2.891*** 0.356 level was selected as the reference age.
Age > 50 1.292** 0.403 Adverse weather 1.740*** 0.250 There was a significant difference visible
between more than 50 years and less than
Driving while fatigued 3.836*** 0.395 Gamma
30 years, but not between 30‒50 years and
Cell phone use 2.988*** 0.489 Speeding 1.128* 0.524 less than 30 years. The at-fault driver’s age
Alcohol use 3.173*** 0.412 Overtaking –2.199** 0.679 of more than 50 years was found to have
significant and intensifying influence on
Speeding 1.409** 0.425 Sharp curve –2.006*** 0.361
the collision severity (estimate = 1.292,
Risky following 2.171*** 0.295 Time of day 24:00–06:00 –1.934*** 0.425 p-value = 0.001). This indicates that older
Dangerous overtaking 2.660*** 0.516 Alpha at-fault drivers are more likely to sustain
more severe crashes, which is consistent
Sharp curve 3.396*** 0.286 α1 –5.756*** 0.568
with previous findings (Zhang et al 2016;
Slippery pavement 1.826*** 0.238 α2 –9.425*** 0.757 Wang et al 2019a). The at-fault driver’s
SeasonsSummer 1.808*** 0.321 Fit-of-goodness age of more than 50 years decreases the
probability of PDO and injury collisions by
Winter 2.481*** 0.328 LL(0) –1 557.06
5.5% and 1.1%, respectively, but increases
Weekends / holidays 0.929** 0.272 LL(β) –474.90 the probability of fatal collision by 6.6%
Pseudo R2 0.695 (see Table 3). A possible explanation is
that truck drivers are more easily fatigued
*p < 0.05; **p < 0.01; ***p < 0.001
while driving on monotonous mountain-
ous expressways for long hours, thus
Table 3 Marginal effects and standard errors of the PCGOL model becoming progressively less sensitive to
Explanatory variable PDO Injury Fatality
emergency conditions.
As expected, the influence of the
Type of collision Head-on –0.053** (0.017) –0.011* (0.005) 0.064** (0.020) at-fault driver’s driving while fatigued
Age More than 50 –0.055** (0.017) –0.011* (0.006) 0.066** (0.021) (est. = 3.836, p-value < 0.001), cell
phone use while driving (est. = 2.988,
Driving while fatigued –0.163*** (0.012) –0.034* (0.013) 0.196*** (0.019)
p-value < 0.001), alcohol-impaired driving
Cell phone use –0.127*** (0.020) –0.026* (0.011) 0.153*** (0.025) (est. = 3.173, p-value < 0.001), and risky
Alcohol use –0.135*** (0.017) –0.028* (0.012) 0.162*** (0.021) following (est. = 2.171, p-value < 0.001)
behaviour has a significantly positive corre-
Speeding –0.060** (0.018) –0.070** (0.025) 0.130*** (0.019)
lation with collision severity. Accordingly,
Risky following –0.092*** (0.012) –0.019* (0.008) 0.111*** (0.016) it can be inferred that at-fault drivers who
Dangerous overtaking –0.113*** (0.021) 0.089** (0.032) 0.024 (0.025) are engaged in these risky types of driving
behaviour are more likely to sustain severe
Sharp curve –0.144*** (0.009) 0.073** (0.023) 0.071** (0.022)
injuries in MME crashes. The marginal
Slippery pavement –0.077*** (0.010) –0.016* (0.006) 0.093*** (0.012) effects analysis also shows that these four
risky driving behaviours significantly
SeasonsSummer –0.077*** (0.013) –0.016* (0.007) 0.093*** (0.017)
enhance the probability of fatal collision
Winter –0.105*** (0.014) –0.022* (0.009) 0.127*** (0.017) but reduce the probability of PDO and
Weekends/holidays –0.039** (0.011) 0.031 (0.017) 0.009 (0.013) injury collisions in MME crashes, as shown
in Table 3. As an example, at-fault driver’s
Time18:00–24:00 –0.031** (0.010) -0.006* (0.003) 0.038** (0.012)
behaviour while driving when fatigued
24:00–06:00 –0.123*** (0.014) 0.074*** (0.019) 0.049*** (0.014) increases the chance of fatal collision by
Adverse weather –0.074*** (0.012) -0.015** (0.005) 0.089*** (0.011) 19.6%, while reducing the chance of PDO
collision by 16.3%, and injury collision by
* p < 0.05; ** p < 0.01; *** p < 0.001
3.4%, respectively. Similar findings have
previously been reported by numerous
collisions, but not between rear-end and of a sideswipe collision, which is in good researchers (Wang & Prato 2019; Wang et
sideswipe collision, or between angle and agreement with previous findings from al 2019a & 2019b; Chen & Zhang 2016).
sideswipe collision. The head-on collision Shaanxi, China (Chen & Zhang 2016). These results strongly suggest that strict
type displayed a significant and positive Specifically, a decrease of 5.3% in PDO laws and regulations should be enforced to
coefficient (estimate = 1.259, p-value = collision and 1.1% in injury collision, and prohibit risky driving behaviour, especially
0.002), indicating that at-fault drivers an increase of 6.4% in fatal collision were for inexperienced and elderly drivers while
involved in a head-on collision are likely observed for MME crashes with head-on navigating sharp curves and steep downhill
to sustain more severe crashes than those collision type. gradients under adverse weather conditions

Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022 67
(i.e. slippery pavement, heavy rain or snow,
and low visibility).
Additionally, at-fault drivers’ speeding
behaviour is shown to have a significant
and pronounced influence on MME crash
severity but violates the proportional odds
assumption. The first panel of coefficient
(i.e. PDO versus injury + fatality) is 1.409
(p-value = 0.001), and the second panel
of coefficient (i.e. PDO + injury versus
fatality) is 2.537; thus it can be concluded
that at-fault drivers who engage in speed-
ing behaviour are likely to sustain more
fatal collisions in MME crashes (Wang &
Prato 2019; Wang et al 2019a; Theofilatos
et al 2018). An increase of 13.0% in fatal
collisions, and a decrease of 6.0% in PDO
collisions and 7.0% in injury collisions were
observed for MME crashes due to at-fault
drivers’ speeding behaviour (see Table 3).
Dangerous overtaking behaviour was also
found to violate the proportional odds
assumption, and the descending series of
coefficients (2.660 versus 0.461) indicated
that at-fault drivers were likely to sustain
more injury collisions, which considerably
altered the probabilities of certain crash
severity (PDO: ‒11.3%; injury: 8.9%; fatal-
ity: 2.4%). This is consistent with previous
reporting (Richter et al 2017). On the other
hand, truck involvement was not found
to show significant influence on MME
crash severity, and a recent examination of
a Greek crash sample also exhibited that
an increased proportion of trucks do not
result in more severe injuries (Theofilatos
et al 2018).
Roadway contributions
Regarding road factors, a sharp curve
violates the proportional odds assumption
and has a significant and positive impact
on MME crash severity p-value < 0.001.
Specifically, the decreasing trend of
panels of coefficient (3.396 versus 0.390)
shows that MME crashes occurring on
sharp curves are more likely to result in
injury collisions, which is consistent with
its marginal effects (see Table 3). These
results correspond well with many previ-
ous findings reported in literature (Meng
2017; Rusli et al 2018; Wang & Prato 2019;
Wang et al 2019a; Yu et al 2015; Chen &
Zhang 2016). Steep slopes, however, were
not found to have significantly correlated
with fatality and injury probabilities in
discordance with previous results (Wang
et al 2019a; Yu et al 2015; Chen & Zhang
2016). The possible reason may be that the
later model structure does not consider
68
the different mechanism between single-
vehicle and multi-vehicle crashes on moun-
tainous expressways. In addition, slippery
pavement conditions have a significantly
positive influence on crash severity (est. =
1.826, p-value < 0.001), increasing the
probability to 9.3% in fatal collisions, while
decreasing the probability to 47.7% and
1.6% in PDO and injury collisions, respec-
tively. Many previous reports in literature
have also confirmed this result (Wang &
Prato 2019; Ma et al 2016; Chen & Zhang
2016).
Environmental conditions
Season is naturally split into four catego-
ries: spring, summer, autumn and winter,
and spring was selected as the reference
category in this study. The modelling
result revealed that there was a significant
difference between summer and spring, as
well as winter and spring, but not between
autumn and spring. Particularly, both
summer (est. = 1.808, p-value < 0.001) and
winter (est. = 2.481, p-value < 0.001) were
significantly and positively correlated with
crash severity, indicating that an MME
crash occurring on a summer or winter’s
day is likely to be a more serious collision
compared to that happening on a spring
day. This considerably alters the prob-
abilities of certain crash severities (PDO:
‒7.7% versus ‒10.5%; injury: ‒1.6% versus
‒2.2%; fatality: 9.3% versus 12.7%), as shown
in Table 3. The possible reason lies in the
adverse effects of rainfall, fog and snowfall
on at-fault driver’s peripheral vision and
vehicle brake performance; however, most
at-fault drivers comprehend the risk of
driving under adverse weather conditions
on summer or winter days, so they may
drive carefully and thus the total number
of crash occurrences could be reduced,
but not the crash severity. This result con-
tradicts our previous finding from Taigan
Expressway, a segment of G45 Daguang
Expressway from Taihe to Ganzhou in
Jiangxi, China (Zhang et al 2016).
Evidently, the period between midnight
and six in the early morning is the most
dangerous time for drivers due to sleepi-
ness or fatigue while driving (Zhang et
al 2016; Meng 2017; Wang & Prato 2019;
Wang et al 2019a; Chen & Zhang 2016) and
violates the proportional odds assumption
with a positive coefficient (est. = 2.891,
p-value < 0.001). As the first panel of coef-
ficient (2.891) is larger than the second
one (0.957), it can be inferred that MME
crashes occurring during the period of
Volume 64 Number 1 March 2022
midnight to 6:00 am are likely to result
in more injury collisions, increasing the
likelihood of injury and fatal collisions by
7.4% and 4.9%, respectively, while decreas-
ing the probability of PDO collision by
12.3%. This is mainly attributed to at-fault
driver sleepiness or fatigue while driving,
as well as darkness or low-light conditions.
During this period, at-fault drivers often
use alcohol, caffeine, or music to keep
themselves awake, but these measures can
significantly distract attention and impair
driving performance (Ronen et al 2014).
Thus, drivers should be educated against
continuous driving while fatigued or sleepy.
Specifically, it is recommended that the
government should formulate strict rules
and regulations to limit maximum nightly
driving hours and minimum rest hours
after continuous or accumulated driving,
especially for those who engage in long-
distance commercial transport, and any
offenders should be seriously punished.
Conversely, the period from 6:00 pm to
midnight does not violate the proportional
odds assumption and is significantly and
positively correlated with crash severity
(est. = 0.733, p-value = 0.001). Also, as illus-
trated in the results of marginal effects (see
Table 3), a reduction of 3.1% in PDO colli-
sion and 0.6% in injury collision, as well as
an increase of 3.8% in fatal collision, were
observed in MME crashes associated with
the period from 6:00 pm to midnight.
Finally, adverse weather conditions (est.
= 1.740, p-value < 0.001) were illustrated
to have a significantly positive association
with collision severity, which indicated that
MME crashes under adverse conditions
(rain, snow or fog) are likely to cause severe
collisions, increasing the chance of fatal
collision by 8.9%, but reducing the possibil-
ity of PDO and injury collisions by 7.4%
and 1.5%, respectively. Obviously, driving
under adverse conditions may increase the
risk of crashes due to the reduced sight
distance, slippery pavement and limited
vehicle manoeuvrability along horizontal
and crest vertical curves on mountainous
expressways. This finding is consistent
with many previous reports (Zhang et al
2016; Meng 2017; Rusli et al 2018: Wang &
Prato 2019; Wang et al 2019a; Yu et al 2015;
Chen & Zhang 2016).
CONCLUSION
This research examined the influence of
potential risk factors on MME crash sever-
ity, as well as the marginal effects of each
Journal of the South African Institution of Civil Engineering
contributory factor, by combining 1 521
multi-vehicle crash samples from one typi-
cal mountainous expressway in Shaanxi,
China, and utilising a partially constrained
generalised ordered logit model. The
statistical results illustrate that fourteen
independent contributory variables had a
significant and intensifying influence on
MME crash severity.
An extremely significant contribu-
tion lies in the findings about the at-fault
driver’s risky driving behaviour on the
fatality probability upon an MME crash
occurrence. There appears to be an
urgent need for enforcement measures to
discourage such risky driving behaviour
like driving while fatigued, cell phone use
while driving, alcohol-impaired driving,
speeding, and risky following, especially
among those engaged in long-distance
commercial transport tasks. Another very
significant contribution lies in the findings
about the effects of roadway geometric
characteristics and environmental condi-
tions. The results also suggest that stricter
police enforcement should be compelling at
slippery and sharp curve segments during
adverse weather and in winter.
This study is not without important
methodological limitations, however.
Firstly, the crash sample was only selected
from one expressway segment in Shaanxi,
China, and may not be representative
of the overall traffic safety situation on
mountainous expressways in the country
as a whole. In addition, the original data
may contain some missing, incomplete, or
possibly incorrect points due to unreported
crashes or injuries and errors involved in
manual data entry. Secondly, a relatively
small sample (1 705 observations) over a
very long period of time (six years) can lead
to unreliable and inaccurate estimations
(Behnood & Mannering 2019), but the
current study does not test the temporal
stability of the estimated parameters over
time. It is worth noting that more recent
data should be collected for the model
estimation in the near future, in which
crash data can be divided over different
time periods and then likelihood ratio
tests will be used to explore whether
temporal instability is an issue. Thirdly, the
data contains a high level of unobserved
heterogeneity, which may affect identify-
ing the exact contributing factors, so the
models accounting for heterogeneity in
data analysis merit further deep investiga-
tion. Finally, the coupling effect of multiple
factors on the severity of MME crashes was
Journal of the South African Institution of Civil Engineering Volume 64 Number 1 March 2022
not focused on in this study, such as the
existence of tunnels and bridges (Sun et al
2020), so one of the greatest future chal-
lenges is to unpack which combinations of
factors produce the greatest risks (Boora et
al 2018; Santos et al 2021), such as driver’s
risk driving behaviour and sharp curves
coincided, and traffic flow and weather
conditions interacted, etc.
It is anticipated that the findings of this
study will provide useful empirical knowl-
edge on the effects of several factors affect-
ing the injury severity of MME crashes.
The proposed probabilistic approach assists
by providing efficient countermeasures
and technical programs for crash preven-
tion and safety performance improvement
on mountainous expressways in Western
China and other countries in the world.
ACKNOWLEDGEMENTS
This work was financially supported by
the Natural Science Foundation of Shaanxi
Province, China, under Grant Number
2020JM-252. The authors acknowledge
the Department of Transport of Shaanxi
Province and Shaanxi Provincial Highway
Bureau for providing crash data.
REFERENCES
Behnood, A & Mannering, F 2019. Time-of-day
variations and temporal instability of factors
affecting injury severities in large-truck crashes.
Analytic Methods in Accident Research, 23: 100102.
https://doi.org/10.1016/j.amar.2019.100102.
Benlagha, N & Charfeddine, L 2020. Risk factors of
road accident severity and the development of a
new system for prevention: New insights from
China. Accident Analysis & Prevention, 136: 105411.
https://doi.org/10.1016/j.aap.2019.105411.
Boora, A, Ghosh, I, Chandra, S & Rani, K 2018.
Measurement of free-flow conditions on multilane
intercity highways under heterogeneous traffic
conditions. Journal of the South African Institution
of Civil Engineering, 60(1): 2–9. https://doi.
org/10.17159/2309-8775/2018/v60n1a1.
Chen, C & Zhang, J 2016. Exploring background risk
factors for fatigue crashes involving truck drivers
on regional roadway networks: A case control study
in Jiangxi and Shaanxi, China. SpringerPlus, 5: 582.
https://doi.org/10.1186/s40064-016-2261-y.
Dong, B, Ma, X, Chen, F & Chen, S 2018. Investigating
the differences of single-vehicle and multivehicle
accident probability using mixed logit model.
Journal of Advanced Transportation, 2018: 2702360.
https://doi.org/10.1155/2018/2702360.
Khan, I U & Vachal, K 2020. Factors affecting
injury severity of single-vehicle rollover crashes
in the United States. Traffic Injury Prevention,
21(1): 66–71. https://doi.org/10.1080/15389588.2
019.1696962.
Ma, X, Chen, S & Chen, F 2016. Correlated
random-effects bivariate Poisson lognormal
model to study single-vehicle and multi-vehicle
crashes. Journal of Transportation Engineering,
142(11): 04016049. https://doi.org/10.1061/(ASCE)
TE.1943-5436.0000882.
Meng, Y 2017. Estimation of crash severity on
mountainous freeways in Chongqing. Mathematical
Problems in Engineering, 2017: 9764309. https://doi.
org/10.1155/2017/9764309.
National Bureau of Statistics of China. China
Statistical Yearbook (2019). http://www.stats.gov.
cn/tjsj/ndsj/2019/indexch.htm.
Peterson, B & Harrel, F E 1990. Partial proportional
odds models for ordinal response variables.
Journal of the Royal Statistical Society: Series C
(Applied Statistics), 39(2): 205–217. https://doi.
org/10.2307/2347760.
Rezapour, M, Moomen, M & Ksaibati, K 2019. Ordered
logistic models of influencing factors on crash
injury severity of single and multiple-vehicle
downgrade crashes: A case study in Wyoming.
Journal of Safety Research, 68: 107–118. https://doi.
org/10.1016/j.jsr.2018.12.006.
Richter, T, Ruhl, S, Ortlepp, J & Bakaba, E 2017. Causes,
consequences and countermeasures of overtaking
accidents on two-lane rural roads. Transportation
Research Procedia, 25: 1989–2001. https://doi.
org/10.1016/j.trpro.2017.05.395.
Ronen, A, Oron-Gilad, T & Gershon, P 2014. The
combination of short rest and energy drink
consumption as fatigue countermeasures during
a prolonged drive of professional truck drivers.
Journal of Safety Research, 49: 39–43. https://doi.
org/10.1016/j.jsr.2014.02.006.
Rusli, R, Haque, M M, King, M & Voon, WS 2017.
Single-vehicle crashes along rural mountainous
highways in Malaysia: An application of random
parameters negative binomial model. Accident
Analysis & Prevention, 102: 153–164. https://doi.
org/10.1016/j.aap.2017.03.002.
Rusli, R, Haque, M M, Afghari, A P & King, M 2018.
Applying a random parameters Negative Binomial
Lindley model to examine multi-vehicle crashes
along rural mountainous highways in Malaysia.
Accident Analysis & Prevention, 119: 80–90. https://
doi.org/10.1016/j.aap.2018.07.006.
Sameen, M I & Pradhan, B 2017. Assessment of the
effects of expressway geometric design features on
the frequency of accident crash rates using high-
resolution laser scanning data and GIS. Geomatics,
Natural Hazards and Risk, 8(2): 733–747. https://
doi.org/10.1080/19475705.2016.1265012.
Santos, K, Dias, J P, Amado, C, Sousa, J & Francisco,
P 2021. Risk factors associated with the increase
of injury severity of powered two wheelers road
accidents victims in Portugal. Traffic Injury
69
 Prevention, 22(8): 646–650. https://doi.org/10.1080
/15389588.2021.1987421.
Sun, Z, Liu, S, Li, D, Tang, B & Fang, S 2020. Crash
analysis of mountainous freeways with high
bridge and tunnel ratios using road scenario-based
discretization. PLoS One, 15(8): e0237408. https://
doi.org/10.1371/journal.pone.0237408.
Theofilatos, A & Ziakopoulos, A 2018. Examining
injury severity of moped and motorcycle occupants
with real-time traffic and weather data. Journal
of Transportation Engineering, Part A: Systems,
144(11): 04018066. https://doi.org/10.1061/
JTEPBS.0000193.
Wang, L, Li, R, Wang, C & Liu, Z 2021a. Driver
injury severity analysis of crashes in a western
China’s rural mountainous county: Taking crash
compatibility difference into consideration. Journal
70
of Traffic and Transportation Engineering (English
Edition), 8(5): 703–714. https://doi.org/10.1016/j.
jtte.2020.12.002.
Wang, Y & Prato, C G 2019. Determinants of injury
severity for truck crashes on mountain expressways
in China: A case-study with a partial proportional
odds model. Safety Science, 117: 100–107. https://
doi.org/10.1016/j.ssci.2019.04.011.
Wang, Y, Luo, Y & Chen, F 2019a. Interpreting risk
factors for truck crash severity on mountainous
freeways in Jiangxi and Shaanxi, China. European
Transport Research Review, 11: 26. https://doi.
org/10.1186/s12544-019-0366-4.
Wang, Y, Li, L & Prato, C G 2019b. The relation
between working conditions, aberrant driving
behaviour and crash propensity among
taxi drivers in China. Accident Analysis &
Volume 64 Number 1 March 2022
Prevention, 126: 17–24. https://doi.org/10.1016/j.
aap.2018.03.028.
Wang Y, Zhang H & Shi N 2021b. Factors contributing
to the severity of heavy truck crashes: A
comparative study of Jiangxi and Shaanxi, China.
Jordan Journal of Civil Engineering, 15(1): 41–51.
Yu, R, Xiong, Y & Abdel-Aty, M 2015. A correlated
random parameter approach to investigate the
effects of weather conditions on crash risk for a
mountainous freeway. Transportation Research,
Part C: Emerging Technologies, 50: 68–77. https://
doi.org/10.1016/j.trc.2014.09.016.
Zhang, C, Guo, X, Wang, Y, Li, Y & Zhang, N 2016.
What leads to severe mountainous freeway crashes
in southeast of China? Technički Vjesnik ‒ Technical
Gazette, 23(6): 1747–1753. https://doi.org/10.17559/
TV-20150516034036.
Journal of the South African Institution of Civil Engineering
Guidelines for the preparation of papers and technical notes
Authors should comply with the following guidelines when preparing papers for publication in the journal
The Journal of the South African Institution of Civil
Engineering is published quarterly in March, June,
September and December. Articles submitted for
publication are reviewed by a panel of referees under
the guidance of the SAICE Journal Editorial Panel.
The journal publishes research papers covering all the
disciplines of civil engineering (structural, geotechnical,
railway, coastal/marine, water, construction, environ-
mental, municipal, transportation) and associated topics
that are relevant to the civil engineering profession, and
that preferably have particular relevance to civil engi-
neering in southern Africa and the African continent.
When preparing articles for publication, authors
should please take note of the following and comply
with the guidelines as set out:
CLASSIFICATION OF ARTICLES
CONSIDERED FOR PUBLICATION
QQ Technical papers are well-researched, in-depth,
fully referenced technical articles not exceeding
6 000 words in length (excluding tables, illustra-
tions 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 technical papers or technical notes on condi-
tion 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 contribute
by synthesising the information and drawing new
conclusions from the dissemination of the previously
published work.
QQ Discussion on published articles is welcomed up to
six months after publication. The length of discus-
sion contributions is limited to 1 500 words. Where
appropriate, discussion contributions will be subject
to the normal reviewing process and will be for-
warded 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 must affirm
that submitted material has not been published
previously, is not under consideration for publica-
tion 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
permission to publish experimental data and other
information that may be confidential or sensitive.
Authors are also responsible for obtaining permis-
sion from copyright owners when reproducing mate-
rial that had been published elsewhere. Proof of such
permission must be supplied.
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 experi-
ence problems uploading your paper, please contact
the 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 docu-
ment. 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 docu-
ment should be typed in Times New Roman 12 pt
font. Contributions should be accompanied 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 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: These
should preferably be submitted in colour, as the
journal is a full-colour publication.
QQ Their positions should be clearly marked in the
text as follows: [Insert Figure 1].
QQ Figures, tables, photos, illustrations and equa-
tions 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).
QQ Only those figures and photographs essential to
the understanding of the text should be included.
All illustrations should be referred to in the text.
QQ Figures should be produced using com-
puter graphics. Hand-drafted figures will not be
­accepted. Lettering on figures should be equiva-
lent 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 graphically.
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 mathemati-
cal/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 preferred.
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) discovered 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 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 the paper will be withdrawn from the system.
FINAL ARTICLE
QQ Copyright: On acceptance of the paper or technical
note, copyright must be transferred by the author/s
to the South African Institution of Civil Engineering
on the form that will be provided by the Institution.
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 Proofs: First proofs of papers will be sent to authors
in PDF format for verification before publication.
No major re-writes will be allowed, only essential
minor corrections.