Volume 63 Number 2

June 2021

QQ  rediction of water reuse potential in South

P
African water management areas: A linear
regression and Bayesian network approach
QQ  self-calibrating model to estimate
A
average speed from AADT
QQ  tatic pile capacity from the rapid test
S
via the unloading-point method
QQ  critical review of the Water Sorptivity Index (WSI)
A
parameter for potential durability assessment: Can
WSI be considered in isolation of porosity?
QQ  outh African infrastructure condition – an opinion
S
survey for the SAICE Infrastructure Report Card
QQ  rediction of California Bearing Ratio and compaction
P
characteristics of Transvaal soils from indicator properties
 Thanks to our Referees!
The SAICE Journal Editorial Panel would like to thank the persons listed below, all of whom served as referees during 2019, 2020 and this year
so far (up to the end of May). The quality of our journal is not only a reflection of the expertise of participating authors, but certainly also of
our referees’ high standards. We sincerely appreciate the supportive spirit in which our referees set aside precious time to perform this task
voluntarily. If we have overlooked anyone by mistake, please accept our apologies, and know that your assistance is equally appreciated!

Prof Mark Alexander Dr Fidelis Emuze Mr Carlo Loubser Dr John Sampson

Dr Joseph Anochie-Boateng Mr Willie Enright Dr Erik Loubser Prof Manu Santhanam
Dr André Archer Prof Phil Everitt Dr Hylton Macdonald Mr Thato Seabi
Mr Preyan Arumugam-Nanoolal Prof George Fanourakis Dr Jeffrey Mahachi Prof Marion Sinclair
Dr Danie Badenhorst Dr William Fedrigo Prof James Maina Dr Alexander Sinske
Mr Allan Bailey Dr Lloyd Fisher-Jeffes Mr Stephen Mallory Dr Sarah Skorpen
Dr Roger Barker Prof Anne Fitchett Prof George Markou Dr Andrew Slaughter
Prof Hans Beushausen Dr Tiago Gaspar Dr Vernon Marshall Prof Jeff Smithers
Mr Keith Bokelman Ms Santie Gouws Mr Adriaan Meintjes Dr Breda Strassheim
Mr Deon Bonthuys Mr Trevor Green Dr Martin Mgangira Mr Hugh Thompson
Mr Dawid Bosman Ms Karien Grove Mr Raimund Miller Dr Luis Torres-Cruz
Mr Ryan Bradley Dr Brian Harrison Dr Terence Milne Dr Anthony Turton
Dr George Brink Mr Makarand Hastak Mr Roelof Minnaar Mr Frans van der Merwe
Mr André Broekman Dr Arno Hefer Prof Mohamed Mostafa Mr Jackie van der Westhuizen
Mr Chris Brooker Dr Francois Heyns Dr Friedel Mulke Mr Marco van Dijk
Ms Megan Bruwer Mr Gerhard Hitge Prof Mike Muller Dr Peter van Niekerk
Prof Walter Burdzik Mr Anton Holtzhausen Mr Marshall Muthen Prof Louis van Rooy
Mr Gys Burger Dr Emile Horak Dr Frank Netterberg Dr Mark van Ryneveld
Dr Nicol Chang Mr Shane Hossell Mr Mike Newham Dr Andre van Tonder
Mr Arthur Chapman Prof Bola Ikotun Ms Chantel Niebuhr Dr Martin van Veelen
Dr Renier Cloete Prof Kim Jenkins Prof Drahomir Novak Prof Fanie van Vuuren
Ms Ashley Coetzee Dr Gary Jones Prof Sunday Nwaubani Ms Sonel van Wageningen
Ms Leanne Coetzee Mr Nico Jonker Mr Johan O’Connell Prof Gideon van Zijl
Dr Louis Coetzee Prof Gerrit Jordaan Mr Patrick Okonkwo Prof Marianne Vanderschuren
Dr James Dabrowski Mr Philip Joubert Prof Mike Otieno Mr Rick Vandoorne
Dr Morris de Beer Ms Janina Kanjee Dr Phil Paige-Green Prof Celeste Viljoen
Dr Nico de Koker Prof Elsabé Kearsley Mr Chris Palmer Prof Alex Visser
Dr Nico-Ben de Villiers Mr John Kempe Mr Michael Palmer Mr Gerrit Visser
Dr Wibke de Villiers Mr Gerhard Keyter Prof Geoff Pegram Dr Jeanine Vonkeman
Dr Erik Denneman Dr Moses Kiliswa Mr Bryan Perrie Dr Eduard Vorster
Prof Dimitris Diamantidis Mr Hans King Dr Pine Pienaar Prof Richard Walls
Prof Mahongo Dithinde Dr Christoff Krogscheepers Mr G V Price Dr Johan Wesseloo
Dr Louw du Plessis Mr Frans Kromhout Prof Kalaga Ramachandra Rao Dr Pieter Wessels
Dr Realize du Plooy Mr Edwin Kruger Dr Hubrecht Ribbens Mr Mike Wiese
Mr Simeon du Preez Prof Samuel Laryea Mr Alan Robinson Prof Jan Wium
Dr Willie Ebersohn Dr Aldu le Grange Prof Chris Roth Prof Alphose Zingoni
Prof Alex Elvin Mr Francois le Roux Dr Martin Rust Prof Mark Zuidgeest
Volume 63 No 2 June 2021 ISSN 1021-2019

Publisher
South African Institution of Civil Engineering
Block 19, Thornhill Office Park,
Bekker Street, 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 Prediction of water reuse potential in South
joint EditorS-in-chief
Prof Chris Clayton African water management areas: A linear
University of Southampton
c.clayton@soton.ac.uk regression and Bayesian network approach
Prof Gerhard Heymann
University of Pretoria S Mamane, A Abimbade, A Ilemobade
gerhard.heymann@up.ac.za
Editor
Verelene de Koker 10 A self-calibrating model to estimate
Tel +27 11 805 5947, Mobile +27 83 378 3996
verelene@saice.org.za
average speed from AADT
journal editorial Panel M M Bruwer
Prof Y Ballim – University of the Witwatersrand
Prof W Burdzik – University of Pretoria
Prof CRI Clayton – University of Southampton
Dr P Day – Jones & Wagener (Pty) Ltd (chairman)
19 Static pile capacity from the rapid test
Prof J du Plessis – University of Stellenbosch
Prof GC Fanourakis – University of Johannesburg
via the unloading-point method
Prof PJ Gräbe – University of Pretoria I Luker
Dr C Herold – Umfula Wempilo Consulting
Prof G Heymann – University of Pretoria
Prof A Ilemobade – University of the Witwatersrand
Prof SW Jacobsz – University of Pretoria 27 A critical review of the Water Sorptivity Index (WSI)
Prof EP Kearsley – University of Pretoria
Prof C Roth – University of Pretoria parameter for potential durability assessment:
Prof W Steyn – University of Pretoria
Mr M van Dijk – University of Pretoria Can WSI be considered in isolation of porosity?
Prof C Venter – University of Pretoria
Prof A Visser – University of Pretoria A J Moore, A T Bakera, M G Alexander
Dr E Vorster – Zutari
Prof J Wium – University of Stellenbosch
Prof A Zingoni – University of Cape Town 35 South African infrastructure condition – an opinion
Prof M Zuidgeest – University of Cape Town
Peer reviewing
survey for the SAICE Infrastructure Report Card
The Journal of the South African Institution of
Civil Engineering is a peer-reviewed journal
F C Rust, K Wall, M A Smit, S Amod
that is distributed internationally
Design and reproduction
Marketing Support Services, Ashlea Gardens, Pretoria
47 Prediction of California Bearing Ratio and
Papers for consideration should be submitted compaction characteristics of Transvaal
online at: http://journal.saice.org.za
soils from indicator properties
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed F J Haupt, F Netterberg
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 Prediction of water reuse
Journal of the South African
Institution of Civil Engineering
potential in South African
ISSN 1021-2019
Vol 63 No 2, June 2021, Pages 2–9, Paper 784 water management areas:
DR SALHA MAMANE is a Senior Lecturer in A linear regression and
Bayesian network approach
the School of Statistics and Actuarial Science
at the University of the Witwatersrand in
Johannesburg. His research interests cover
both theoretical and applied statistics. The
general theme of his research activity is
modelling the interrelations of variables in a
S Mamane, A Abimbade, A Ilemobade
complex or dynamical system.

Contact details:
School of Statistics and Actuarial Science
University of the Witwatersrand
Increasing scarcity of freshwater resources and growing environmental awareness have given rise
Private Bag 3 to the use of treated wastewater as an additional source of water supply. However, the amount
Wits 2050 of wastewater that can be reclaimed for reuse is subject to many factors, ranging from technical
South Africa
to socio-economic, environmental and institutional. This paper presents two prediction models
T: +27 11 717 6090
E: salha.mamane@wits.ac.za for wastewater reuse potential in South Africa. The proposed models, namely a Bayesian Network
model and a Linear Regression model, were used as a platform for integrated analysis of features
DR ABIOLA ABIMBADE (AMSAICE, MWISA, such as water use, and socio-demographic and environmental factors. A dataset from the sub-
MNSE) is a Civil Engineer at Lidwala regions of the 19 water management areas in South Africa was employed as input into the models
Consulting Engineers where he consults on
in order to analyse these features and to consequently predict water reuse potential.
turnkey solutions and projects, design of
bulk water and sanitation infrastructure,
and township development and
establishment in southern Africa. His INTRODUCTION Service 2014) and Western Cape. The dams
research interests include water reuse
Water reuse has attracted growing atten- in the North West and Free State provinces
sustainability, decision support tools, water services management, and
legislative and institutional arrangement in the water sector. He is actively tion across the globe as a vital part of water recorded low storage and this was an indica-
involved in research and consulting activities. resources management and forming an tion of deficient rainfall that had led to dry
Contact details: essential component of water conservation. conditions (Department of Water Affairs SA
Lidwala Consulting Engineers Water scarcity challenges, linked with 2014). In 2015–2016, South Africa’s agricul-
2 Payne Road increasing water demand from an increasing tural activities were threatened by a severe
Corner Main Office Park
Bryanston 2021
population, climate change and depletion drought which was caused by an El Nino
South Africa of groundwater resources among other fac- weather system that swept across southern
T: +27 11 793 5486 tors, drive the need for exploring alternative Africa (Baudoin et al 2017). In 2020, Botai
E: aabimbade@lidwala.com / abiolaabim@gmail.com
water resources (Chen et al 2012). With et al (2020) reported that five of the most
increasing knowledge and understanding economically active provinces of South
PROF ADESHOLA ILEMOBADE (Pr Eng, MAP,
MSAICE, MWISA) is an Associate Professor
of the benefits of water reuse, some direct Africa were recovering from the severe
at the School of Civil and Environmental and indirect potable water reuse schemes drought, which had caused negative socio-
Engineering, University of the have been successfully implemented in economic impacts. The adverse climatic
Witwatersrand, Johannesburg, where he
both developed and developing countries conditions resulted in the critical shortage
teaches infrastructure planning and water
engineering subjects at undergraduate and (Leverenz et al 2011; Ilemobade et al 2013). of potable water and damages to crops, to
postgraduate levels. His research interests Many studies on drought in South Africa the point that the North West Province was
are in the areas of water reuse, water demand, water conservation and have alluded to the occurrence of widespread declared a state of disaster (Botai et al 2016).
water supply. He has published extensively, received awards for research
and persistent drying conditions in several The Western Cape was the most adversely
conducted and supervised, and led several multi-disciplinary research
and consulting teams. parts of the country (Botai et al 2016; Du affected (Archer et al 2019). These drought-
Plessis & Schloms 2017; Botai et al 2018; related impacts re-emphasise the need to
Contact details:
School of Civil and Environmental Engineering Archer et al 2019). According to Swemmer increase efforts towards the development of
University of the Witwatersrand et al (2018), northeastern Mpumalanga, the alternative water supply sources. According
Private Bag 3 eastern half of the Free State and northern to Ilemobade et al (2013), the implementa-
Wits 2050
South Africa
KwaZulu-Natal experienced more severe tion of water reuse schemes to mitigate
T: +27 11 717 7153 droughts than in the previous years, as indi- water deficits has increased in many South
E: adesola.ilemobade@wits.ac.za cated by recorded Standardised Precipitation African communities.
Index values for South Africa over a two- The quantity of usable return flow that
year period, as of May 2016. These condi- can be harnessed for reuse is subject to
tions were also noticed in the drought-prone several factors, ranging from environmental
Keywords: water reuse, wastewater recycling, Bayesian network, areas of the North West Province, Northern to technical, socio-economical and institu-
linear regression Cape Province (South African Weather tional (Yang et al 2007). Therefore, several

Mamane S, Abimbade A, Ilemobade A. Prediction of water reuse potential in South African water management areas: A linear regression and
2 Bayesian network approach. J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #784, 8 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a1
researchers have used different decision sup-
port tools for estimating water reuse poten-
tial for specific contexts. These decision sup-
port tools include linear programming (Chu
et al 2004; Yang & Abbaspour 2007; Goyal
& Kumar 2020), mass balance (Hochstrat et
al 2005), ratio analysis (Alfarra et al 2011),
surveys (Adewumi et al 2010), fuzzy infer-
ence (Almeida et al 2013), aggregated index
(Shafiquzzaman et al 2018) and analytic
hierarchy process (Fukasawa & Mierzwa
2020). In this study, a Bayesian Network and
a Linear Regression model were employed.
In the field of natural resources manage-
ment and environmental sciences, Bayesian
Networks have proved to be an appropriate
approach for ecosystem modelling (Rositano
& Ferraro 2014; Garcia-Herrero et al 2013),
climate change impact assessment (Richards
et al 2013; Mantyka-Pringle et al 2014)
and watershed management (Barton et al
2008; Keshtkar et al 2013; Shenton et al
2014). According to Mamitimin et al (2015),
Bayesian Networks have the ability to clearly
explain intricate relations between variables,
easily compare scenarios, and determine the
critical influencing system variables when
applied to natural resources management.
Another key benefit of this approach is
its flexibility with regard to data sources.
In situations where data is limited or not
available, Bayesian Networks can implicitly
integrate relevant data from different sources
such as literature, empirical data, expert and
stakeholders’ knowledge (Uusitalo 2007; Gret-
Regamey et al 2013). This study contributes
to water reuse research by using water supply,
and socio-demographic and environmental
factors to estimate water reuse potential in
the South African water management areas.
METHODOLOGY
The estimation of usable return flow can be
an arduous task where accurate volumetric
information is not readily available, as is
the case in most developing countries. This
problem is further amplified by the follow-
ing two (of several) reasons highlighted by
Grobicki and Cohen (1999) in their study on
water reuse projects in South Africa:
1. In a situation where information does
exist, the data is often found to be
incomplete and tedious to collate. In
some instances, information only exists
in handwritten format and this makes it
difficult to access and analyse.
2. There are often contradictions in
numerical data obtained from different
sources for the same area.
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
65
60
55
50
45
Freshwater use (%)
40
35
30
25
20
15
10
5 3
0
Afforestation
Figure 1 W
 ater use in South Africa for the year 2000
The study by Carden (2013) also corroborat-
ed the above data challenges in South Africa.
In this study, data gathered from sources
such as the Department of Water Affairs,
Statistics South Africa and the South
African Weather Service was used. Water
abstracted from a surface or groundwater
source is said to be consumptively used
when it is no longer available for beneficial
use because it has evaporated, transpired,
been incorporated into products and crops,
consumed by man or livestock, or otherwise
removed from freshwater resources. The
non-consumptive water use yields wastewa-
ter that serves the purpose of contributing
to ecological water reserve and usable return
flow. The reuse of this usable yield may,
however, be further constrained by a variety
of pollution sources, thereby making it chal-
lenging to estimate the portion of this yield
that is available for reuse. Figure 1 shows
that 85% of overall freshwater use in South
Africa is for the agricultural and urban
sectors. Twenty years after publication,
this proportion of freshwater use has not
changed significantly. Therefore, the models
proposed in this paper focus on predicting
water reuse potential in the agricultural and
urban sectors. Figure 2 outlines the pro-
cesses involved in model development.
Data description
The variables used in this study are
described below:
QQ Water use (WU): Refers to the water
abstracted from surface and groundwa-
ter for agricultural, domestic and other
use that sustains life, industrial and
other anthropogenic activities (million
m3 per annum).
62
23
8
4
Agricultural Mining and Rural Urban
industrial
Water use sectors
QQ Agricultural water use (AWU): Refers
to water used for irrigation and other
agricultural activities in the agriculture
sector (million m3 per annum).
QQ Domestic water use (DWU): Refers to
water used for domestic activities such
as cooking, drinking, bathing, laundry
and light industrial activities in the
urban sector (million m3 per annum).
QQ Cost of irrigation water (CIW): (cents of
Rand/m3).
QQ Cost of domestic water (CDW): (cents of
Rand/m3).
QQ Standard precipitation index (SPI):
This is a tool used by the South Africa
Weather Services for measuring the
severity of drought based on rainfall
data.
QQ Water stress indicator (WSI): This is a
function of water availability and water
use (Smakhtin et al 2005). This indica-
tor is defined by the ratio:
FreshwaterWithdrawals
WSI =
(Meanannualrunoff –
Ecologicalwaterreserve)
QQ Population density (PD): This is the
number of inhabitants per unit area
(population/km2).
QQ Water volume per person per day
(WVPP): The average volume of water
used per individual in a day.
QQ Percentage of non-revenue water (NRW):
This refers to all the water that is lost
through physical leakage or commercial
losses (meter under-registration, billing
errors, theft, etc), as well as any unbilled
authorised consumption (fire-fighting,
mains flushing, etc).
3

WU, AWU, CIW,
WSI, SPI
Logarithm Transformation
Regression
Bayesian
network
WU, AWU,
AWU
CIW
URFA URFA
Figure 2 S chematic representation of the processes involved in model development
QQ Percentage of flush toilets (FT) con-
nected to sewerage system.
QQ Usable return flow from agricultural
water (URFA): Volume of reusable water
from agricultural activities (million m3
per annum).
QQ Usable return flow from domestic
(urban) water (URFU): Volume of reus-
able water from domestic activities
(million m3 per annum)
The data that was gathered came from
82 areas. Some areas had both domestic
(urban) and agricultural water use, while
some only had one water use. The Water
Management Areas with no water use
recorded in the sector or no usable return
flow from the sector were discarded from
the analysis. Furthermore, for the domestic
sector, the data was further restricted to
areas with annual domestic water use less
than 50 million m3. This restriction was
implemented to arrive at data that was
more homogeneous, and this helped to
improve the predictions. As a result, the
data that was analysed consisted of 62
observations for the agricultural sector and
44 observations for the domestic (urban)
sector. For the prediction of usable return
flow from the agricultural sector, the
predictor variables that were considered
4 Volume 63 Number 2 June 2021
QQ Bayesian Network model: A graphical
model (Koller & Friedman 2009) for
WU, DWU, CDW,
Input a collection of random variables is a
WVPP, NRW, FT,
variables
PD, WSI family of probability distributions such
that each satisfies a set of conditional
independence relations encoded in a
graph. This provides a parsimonious
parameter and modular representation
Logarithm of the joint distribution of the random
variables in the model, thereby allowing
estimation of model parameters with a
reasonable amount of data and a more
effective computation of marginal
posterior distributions. A Bayesian
Model Regression
Bayesian Network (Koller & Friedman 2009) is a
network
graphical model wherein the conditional
independence relations are encoded in
a directed acyclic graph (a graph with
orientated edges and such that there are
no paths from a node and back).
Variable WU, DWU,
DWU For each sector, the graph that represents the
selection PD
dependence structure of the variables is learnt
from the data. It corresponds to the structure
that maximises the Bayesian Information
Criterion (Konishi & Kitagawa 2008). For a
Prediction URFU URFU more robust estimation of the structure, the
average of graphs learnt from 1 000 bootstrap
samples was used. The optimisation was
carried out using the hill-climbing algorithm
(Nagarajan et al 2013). Thereafter, maximum
were WU, AWU, CIW, WSI and SPI. For likelihood estimates of the parameters of the
the prediction of usable return flow from Gaussian Bayesian Network were calculated.
the domestic sector, the predictor variables
considered were WU, DWU, CDW, WVPP,
NRW, FT, WSI, PD and SPI. SOFTWARE
In this study, the R statistical software
Description of the models (RDevelopment Core Team 2017) was used
For each sector considered (agricultural for the data analysis. The bnlearn package
and domestic), two models were explored, (Scutari 2009) was used for the Bayesian
namely a Linear Regression model and Network modelling.
a Bayesian Network model. Since the
variables were right-skewed and presented
some outliers, a log transformation of the RESULTS
variables was undertaken, which resulted
in better prediction accuracy. The smear- Usable return from the
ing estimate of bias was used to correct for agricultural sector
the bias in the retransformed predictions Table 1 summarises the distribution of the
(Newman 1993). percentage of agricultural water use that is
QQ Linear Regression model: The usable potentially reusable. In particular it indi-
return flow was expressed as a linear cates that, on average, 8% of agricultural
function of the other variables. The water use is potentially reusable.
selection of the variables included in As shown in Figure 3, URFA is highly
the model was performed using the correlated with AWU (0.95) and WU (0.76).
Akaike Information Criterion (Konishi The latter variables are also highly correlated
& Kitagawa 2008). (0.79). Albeit very weak, the correlations of
Table 1 Summary of reusable flow percentage from agricultural water use in South Africa
Minimum 1st Quartile Median Mean 3rd Quartile Maximum
2 5 10 8 10 17
Journal of the South African Institution of Civil Engineering
 CIW and SPI with WU and AWU, and hence
with URFA, are negative, as expected. The

WU
obtained strong linear correlation between
URFA and AWU already signifies that more
WU 1.00

AWU
than 90% of the variability of URFA can be
explained by the variability of AWU.
AWU 0.79 1.00

CIW
Linear Regression model for usable
return flow from agricultural water use
CIW –0.17 –0.32 1.00 In the sequel, N(µ, σ) denotes the normal dis-

WSI
tribution with mean µ and standard deviation
σ and by abuse of notation, a random obser-
WSI 0.33 0.39 –0.20 1.00 vation from the said probability distribution.

SPI
For the usable return flow from
agricultural water use, the fitted Linear

URFA
SPI –0.20 –0.10 0.05 –0.10 1.00
Regression model is:

ln(1+URFA) = 0 .101 ln(1+AWU) +

URFA 0.76 0.95 –0.22 0.32 –0.13 1.00
N(0, 0.075) (1)

Equation 1 implies that when AWU

–1.00 –0.67 –0.33 0 0.33 0.67 1.00 doubles, URFA increases on average by
around 7%.
Figure 3 P
 earson correlation matrix of the variables impacting usable return flow from agricultural
water use Performance of the Linear Regression
model on cross-validation data:
Table 2 summarises the absolute percent-
120 age prediction errors using the Linear
Regression model (Equation 1). It shows in
110 particular that about half of the predictions
were less than 24% higher or lower than the
100
actual observation.
Figure 4 shows the actual and pre-
90
dicted usable return flow from agricultural
Predicted URFA (millions m3 per annum)

water use.
80

70 Bayesian Network model for usable

return flow from agricultural water use
60 Figure 5 shows the structure of the
Bayesian Network learnt from the data.
50

40
WU CIW
30

20

10
AWU
0
0 10 20 30 40 50 60 70 80 90 100 110 120
Observed URFA (millions m3 per annum)

Figure 4 A
 ctual versus predicted usable return flow from agricultural water use using the
Linear model URFA

Table 2 S ummary of absolute percentage prediction errors (agricultural water, Linear Regression model)
Minimum 1st Quartile Median Mean 3rd Quartile Maximum Figure 5 B
 ayesian Network representation of
factors impacting usable return flow
1 15 24 39 46 322
from agricultural water use

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 5
Only the variables WU, CIW and AWU
were found to be factors significantly 100
impacting URFA.
For URFA, the fitted Gaussian Bayesian 90
Network model is the following:
80
ln(1+URFA) = 0 .101 ln(1+AWU) +
N(0, 0.075) (2)

Predicted URFA (millions m3 per annum)

70

ln(1+AWU) = 0 .981 ln(1+WU) − 0.508

ln(1+CIW) + N(0, 0.6) (3) 60

ln(1+WU) = N(4.855, 0.866) (4) 50

ln(1+CIW) = N(1.030, 0.204) (5)

40

30
Performance of the Bayesian Network
model on cross-validation data:
Table 3 summarises the absolute percent- 20
age prediction errors achieved by the
Bayesian Network model. In particular 10
it indicates that about half of the pre-
dictions are within 25% of the actual
0
observations. 0 10 20 30 40 50 60 70 80 90 100
Figure 6 shows the actual versus pre- Observed URFA (millions m3 per annum)
dicted usable return flow from agricultural
water use using the Bayesian Network Figure 6 A
 ctual versus predicted usable return flow from agricultural water use using the
model. Bayesian Network model

Usable return flow from Table 3 S ummary of absolute percentage prediction errors (agricultural water, Bayesian Network model)
domestic activities Minimum 1st Quartile Median Mean 3rd Quartile Maximum
Table 4 summarises the distribution of
2 18 25 39 47 313
the percentage of domestic water used
that is potentially reusable. It shows that,
on average, 34% of domestic water used is
potentially reusable.
WU

As shown in Figure 7, URFU is

DWU

highly correlated with DWU (0.89) and WU 1.00

moderately correlated with WU (0.54)
WVPP

DWU 0.58 1.00

and PD (0.56). CDW is poorly, but expect-
edly negatively, correlated with WU and
NRW

WVPP 0.31 0.30 1.00

WVPP. The obtained strong linear cor-
relation between URFU and DWU already NRW
0.05 0.17 0.39 1.00
PD

indicates that about 79% of the variability

PD
of URFU can be explained by the vari- 0.44 0.73 0.33 0.11 1.00
FT

ability of DWU. FT
CDW

0.27 0.31 0.22 0.07 0.25 1.00

Linear Regression model for usable CDW
WSI

–0.10 0.05 –0.26 –0.37 0.11 0.02 1.00

return flow from domestic water use
WSI
For the usable return flow from domestic
0.63 0.11 0.12 0.03 0.15 0.08 –0.07 1.00
SPI

water use, the fitted Linear Regression SPI

URFU

model is: –0.39 –0.29 –0.09 –0.03 –0.44 0.11 –0.03 –0.13 1.00
URFU
ln(1+URFU) = 0 .612 ln(1+DWU) + 0.54 0.89 0.27 0.15 0.56 0.28 –0.04 0.16 –0.28 1.00
N(0, 0.390) (6)

Equation 6 implies that when DWU –1.00 –0.67 –0.33 0 0.33 0.67 1.00
doubles, URFU increases on average by
around 50%. Figure 7 P
 earson correlation matrix of factors impacting usable return flow from domestic water use

6 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
Table 4 Summary of reusable flow percentages from domestic water use in South Africa
Minimum 1st Quartile Median Mean 3rd Quartile Maximum PD
3 29 36 34 43 59

16

DWU URFU

14

12
Predicted URFU (millions m3 per annum)

WU
10

Figure 9 B
 ayesian Network representation of
8 factors impacting the usable return
flow from domestic water use

6 Figure 10 shows the actual and predict-

ed usable return flow from domestic water
use using the Bayesian Network model.
4

DISCUSSION AND CONCLUSION

2
The development of decision support
models that utilise demographic, water and
environment-related factors to predict usa-
0
0 2 4 6 8 10 12 14 16 ble return flow can be of help to decision-
Observed URFU (millions m3 per annum) makers and can make discussions about
water reuse more efficient in communities
Figure 8 A
 ctual versus predicted usable return flow from urban water use using the Linear model faced with water scarcity. Information on
usable return flow is imperative to the
Performance of the Linear Regression ln(1+DWU) = 0 .259 ln(1+WU) + 0.351 development and implementation of water
model on cross-validation data: ln(1+PD) + N(0, 0.667) (8) reuse initiatives, as well as provide warn-
Table 5 summarises the absolute percentage ing signals that can predict water supply
prediction errors achieved by the Linear ln(1+WU) = N(4.62, 0.91) (9) shortages. In the agricultural sector, the
Regression model (Equation 6). In particular variables that were found to be most useful
it shows that about half of the predictions ln(1+PD) = N(3.576, 1.185) (10) for the prediction of usable return flow
are within 23% of the actual observation. were the total water use, the agricultural
Figure 8 shows the predicted and actual water use and the cost of irrigation water.
usable return flow from urban water use. Performance of the Bayesian In the urban (domestic) sector, the vari-
It can be observed that the model tends to Network on cross-validation data: ables that were found to be the most useful
underestimate higher usable return flow Table 6 summarises the percentage pre- for the prediction of usable return flow
from urban water use. diction errors achieved by the Bayesian were the total water use, the domestic
Network model. In particular it indicates water use and the population density. For
Bayesian network model for usable that about half of the predictions are more the agricultural sector, the Linear and the
return flow from domestic water use than 23% higher or lower than the actual Bayesian Network models yielded a median
Figure 9 shows the structure of the observation. percentage absolute error of 24% and 25%
Bayesian Network learnt from the data.
The variables WU, DWU and PD are found Table 5 S ummary of absolute percentage prediction errors (domestic water, Linear Regression model)
to be significant factors impacting the Minimum 1st Quartile Median Mean 3rd Quartile Maximum
usable return flow from urban water.
1 11 23 56 46 789
The fitted Bayesian Network model for
the prediction of the usable return flow
from domestic water use is: Table 6 S ummary of absolute percentage prediction errors (urban water, Bayesian Network model)
Minimum 1st Quartile Median Mean 3rd Quartile Maximum
ln(1+URFU) = 0 .612 ln(1+DWU) +
1 10 23 58 45 892
N(0, 0.390) (7)

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 7
 of squared Mahalanobis distances with
16 corresponding expected chi-square values
(Figures 11 and 12). In summary, the study
shows that 8% of the agricultural water
14
use is potentially reusable, while 34% of
total domestic water use is potentially
reusable. The study also shows that, given
Predicted URFU (millions m3 per annum)

12
the sectorial water use in a region, the
usable return flow from the sector can
10
be predicted with a reasonable degree of
accuracy. Since the recorded usable return
8 flows are constrained by the availability
of infrastructure, regions with very low
observed return flows, compared to the
6 predicted ones, will tend to be regions less
endowed with the necessary infrastructure.
4 The result of this study also encourages
the discussion of challenges facing reliable
water supply, taking into consideration the
2 effect of climate change.

0
0 2 4 6 8 10 12 14 16 REFERENCES
Observed URFU (millions m3 per annum) Adewumi, J, Ilemobade, A & Van Zyl, J 2010. Decision
support for the planning of integrated wastewater
Figure 10 A
 ctual versus predicted usable return flow from urban water use using the Bayesian reuse projects in South Africa. Water Science and
Network model Technology: Water Supply, 10(2): 251–267.
Alfarra, A, Kemp-Benedict, E, Hotzl, H, Sader, N &
Sonneveld, B 2011. A framework for wastewater reuse
Q–Q plot of Mahalanobis D2 vs quantiles of χ2nvar in Jordan: Utilizing a modified wastewater reuse index.
Water Resources Management, 25(4): 1153–1167.

35 Almeida, G, Vieira, J, Marques, A S, Kiperstok, A

& Cardoso, A 2013. Estimating the potential

40 water reuse based on fuzzy reasoning. Journal of

Environmental Management, 128: 883–892.
Archer, E, Landman, W, Malherbe, J, Tadross, M &

47 Pretorius, S 2019. South Africa’s winter rainfall

30
region drought: A region in transition. Climate Risk
30
Management, 25: 100188.
Barton, D, Saloranta, T, Moe, S, Eggestad, H & Kuikka,
Mahalanobis D2

S 2008. Bayesian belief networks as a meta-modelling

tool in integrated river basin management pros and
20
cons in evaluating nutrient abatement decisions

46 under uncertainty in a Norwegian river basin.

8 37 Ecological Economics, 66(1): 91–104.

4544 Baudoin, M, Vogel, C, Nortjea, K & Naika, M 2017. Living

10
48 with drought in South Africa: Lessons learnt from the
54
recent El Niño drought period. International Journal
42 of Disaster Risk Reduction, 23: 128–137.
Botai, C M, Botai, J O & Adeola, AM 2018. Spatial
0
distribution of temporal precipitation contrasts in
0 5 10 15 South Africa. South African Journal of Science, 114
Quantiles of χ2nvar (7/8). doi:10.17159/sajs.2018/20170391.
Botai, C M, Botai, J O, Adeola, A M, De Wit, J P,
Figure 11 M
 ultivariate outliers in the case of URFA prediction Ncongwane, K P & Zwane, N N 2020. Drought risk
analysis in the Eastern Cape Province of South
respectively. For the domestic sector, the the domestic sector. The performance of Africa. The Copula Lens Water, 12: 1938. doi:
median percentage absolute error deter- the prediction models was certainly nega- 10.3390/w12071938.
mined for both sectors was 23%. The pre- tively affected by the presence of multivari- Botai, C M, Botai, J O, Dlamini, L, Zwane, N & Phaduli,
dictions were generally more accurate for ate outliers, as indicated by a comparison E 2016. Characteristics of droughts in South

8 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering

Q–Q plot of Mahalanobis D2 vs quantiles of χ2nvar
40
30
Mahalanobis D2
20
10
0
5 10
Quantiles of χ2nvar
Figure 12 M
 ultivariate outliers in the case of URFU prediction
Africa: A case study of Free State and North West
provinces. Water, 8: 439.
Carden, K 2013. A measure of sustainability in the
context of urban water management in South Africa.
PhD thesis, University of Cape Town.
Chen, Z, Ngo, H H & Guo, W 2012. A critical review on
sustainability assessment of recycled water schemes.
Science of the Total Environment, 426: 13–31.
Chu, J, Chen, J, Wang, C & Fu, P 2004. Wastewater
reuse potential analysis: Implications for China’s
water resources management. Water Research,
38(11): 2746–2756.
Department of Water Affairs (South Africa) 2014.
Annual National State of Water Resources Report
for the Hydrological Year 2012/2013. https://www.
gov.za.documents (accessed on 14 July 2015).
Du Plessis, J A & Schloms, B 2017. An investigation
into the evidence of seasonal rainfall pattern shifts
in the Western Cape, South Africa. Journal of the
South African Institution of Civil Engineering,
59(4): 47–55.
Fukasawa, B N & Mierzwa, J C 2020. Identification of
water reuse potential in metropolitan regions using
the analytic hierarchy process. Environmental and
Sustainability Indicators, 8: 100064.
Garcia-Herrero, S, Mariscal, M, Gutierrez, J & Toca-
Otero, A 2013. Bayesian network analysis of safety
culture and organizational culture in a nuclear
power plant. Safety Science, 53:82–95.
Goyal, K & Kumar, A 2020. A modelling approach to
assess wastewater reuse potential for Delhi city.
Water Supply, 20(5): 1716–1725.
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
26
38
8 Journal of Applied Ecology, 51(3): 572–581.
32
52 39 53
10
30 6
14
15 20
Gret-Regamey, A, Brunner, S H, Altwegg, J &
Bebi, P 2013. Facing uncertainty in ecosystem
services-based resource management. Journal of
Environmental Management, 127: S145–S154.
Grobicki, A & Cohen, B 1999. Water reclamation for
direct re-use in urban and industrial applications
in South Africa and its projected impact upon water
demand. Pretoria: Water Research Commission.
Hochstrat, R, Wintgens, T, Melin T & Jeffrey, P 2005.
Wastewater reclamation and reuse in Europe: A
model-based potential estimation. Water Science
and Technology: Water Supply, 5(1): 67–75.
Ilemobade, A, Olanrewaju, O & Griffioen, M 2013.
Greywater reuse for toilet flushing at a university
academic and residential building. Water SA,
39(3): 351–360.
Keshtkar, A, Salajegheh, A, Sadoddin, A & Allan,
M G 2013. Application of Bayesian networks for
sustainability assessment in catchment modeling
and management (Case study: The Hablehrood
River catchment). Ecological Modelling, 268: 48–54.
Koller, D & Friedman, N 2009. Probabilistic Graphical
Models: Principles and Techniques. Cambridge, MA:
MIT Press.
Konishi, S & Kitagawa, G 2008. Bayesian Information
Criteria and Statistical Modeling. New York: Springer.
Leverenz, H L, Tchobanoglous, G & Asano, T 2011.
Direct potable reuse: A future imperative. Journal of
Water Reuse and Desalination, 1(1): 2–10.
Mamitimin, Y, Feike, T & Doluschitz, R 2015. Bayesian
network modeling to improve water pricing practices
in northwest China. Water, 7(10): 5617–5637.
Mantyka-Pringle, C S, Martin, T G, Moffatt, D B, Linke,
S & Rhodes, J R 2014. Understanding and predicting
the combined effects of climate change and land-use
change on freshwater macroinvertebrates and fish.
Nagarajan, R, Scutari, M & Lebre, S 2013. Bayesian
Networks in R with Applications in Systems Biology.
New York: Springer Science+Business Media.
Newman, M C 1993. Regression analysis of log-
transformed data: Statistical bias and its correction.
Environmental Toxicology and Chemistry,
12(6): 1129–1133.
R Development Core Team 2017. A language and
environment for statistical computing. Vienna,
Austria. https://www.r-project.org.
Richards, R, Sano, M, Roiko, A, Carter, R, Bussey, M,
Matthews, J & Smith, T F 2013. Bayesian belief
modeling of climate change impacts for informing
regional adaptation options. Environmental
Modelling & Software, 44: 113–121.
Rositano, F & Ferraro, D O 2014. Ecosystem services
provided by agroecosystems: A qualitative and
quantitative assessment of this relationship in
the La Pampa region, Argentina. Environmental
Management, 53(3): 606–619.
Scutari, M 2009. Learning Bayesian networks with the
bnlearn R package. arXiv preprint arXiv:09083817.
https://arxiv.org.
Shafiquzzaman, M, Haider, H, AlSaleem, S S, Ghumman,
A R & Sadiq, R 2018. Development of Consumer
Perception Index for assessing greywater reuse potential
in arid environments. Water SA, 44(4): 771–781.
Shenton, W, Hart, B T & Chan, TU 2014. A Bayesian
network approach to support environmental flow
restoration decisions in the Yarra River, Australia.
Stochastic Environmental Research and Risk
Assessment, 28(1): 57–65.
Smakhtin, V, Revenga, C & Döll, P 2005. Taking
into account environmental water requirements
in global-scale water resources assessments.
International Water Management Institute.
https://core.ac.uk/download/pdf/6405183.pdf.
South African Weather Service 2014. Annual Report
2013/2014. https://www.saws.co.za (accessed on
10 August 2015).
Swemmer, A M, Bond, W J, Donaldson, J, Hempson,
G P, Malherbe, J & Smit, I P J 2018. The ecology of
drought – A workshop report. South African Journal
of Science, 2018: 114(9/10). https://doi. org/10.17159/
sajs.2018/5098.
Uusitalo, L 2007. Advantages and challenges of
Bayesian networks in environmental modelling.
Ecological Modelling, 203(3–4): 312–318.
Yang, H & Abbaspour, K C 2007. Analysis of
wastewater reuse potential in Beijing. Desalination,
212(1–3): 238–250.
Yang, J, Reichert, P & Abbaspour, K C 2007. Bayesian
uncertainty analysis in distributed hydrologic
modeling: A case study in the Thur River basin
(Switzerland). Water Resources Research, 43(10).
9
TECHNICAL PAPER
A self-calibrating model
to estimate average
Journal of the South African
Institution of Civil Engineering

speed from AADT

ISSN 1021-2019
Vol 63 No 2, June 2021, Pages 10–18, Paper 953

MEGAN BRUWER (AMSAICE) is a

transportation engineer. She joined the Civil M M Bruwer
Engineering Department of Stellenbosch
University in 2015 as a lecturer, and as the
project coordinator of the Stellenbosch
Smart Mobility Laboratory (SSML). Prior to Transport practitioners need a universally applicable speed prediction model to estimate
joining Stellenbosch University she worked
average speeds on any road. Average annual speed is a key input to the economic assessment
as a transport engineering consultant,
where she was involved in the implementation and operational design of of transport infrastructure where reliable estimates of future average speeds are necessary to
public transport systems and road-based traffic accommodation for new calculate economic costs and benefits. The relationship between Annual Average Daily Traffic
developments. Her research interests include traffic-flow theory and the (AADT) and average annual speed was investigated on higher-order roads across South Africa,
application of Intelligent Transport Systems to improve traffic data
revealing a high level of variability in this correlation at different locations. This variation is
collection for transport planning and traffic management. She is currently
completing a PhD in this field. influenced by road characteristics, such as alignment and cross-section, complicating the
formulation of a universal speed prediction model. Two novel speed prediction models are
Contact details:
Lecturer and Project Coordinator: proposed in this article that use AADT to forecast future average annual speed. The speeds
Stellenbosch Smart Mobility Lab of heavy vehicles and light vehicles can be estimated separately, as well as the average speed
Department of Civil Engineering of all vehicles simultaneously. Both models are self-calibrating, accounting for the variation
Stellenbosch University
Private Bag X1
in the AADT–speed relationship. This calibration step is unique to speed prediction models
Matieland and increases the reliability of these models to estimate future average speeds considerably.
Stellenbosch 7602 Furthermore, self-calibrating average annual speed prediction models are universally applicable
South Africa
and will simplify economic assessment of transport infrastructure.
T: +27 21 808 4080
E: mbruwer@sun.ac.za

Introduction certain road and depends on surrounding

Keywords: speed prediction, average annual speed, self-calibration,
AADT, economic assessment
New transport infrastructure is imple-
mented only if it offers economically
beneficial solutions to existing needs. The
economic assessment of transport facili-
ties requires reliable estimation of average
annual speeds at future intervals to calcu-
late economic costs and benefits. Currently,
the few South African methods that exist
to estimate the progression of average
annual speeds are determined for – and are
therefore only applicable to – particular
roads. There is a need for a generic model
to predict future speeds on higher-order
roads for use in economic evaluation.
The annual progression of traffic
volume is generally agreed upon, i.e. we
anticipate an increase in traffic year-on-
year, assumed to be related to population
growth and development. The rate at
which traffic increases annually is typically
modelled using compound growth based
on the economic principle of compound
interest. In South Africa, annual traffic
growth of below 3% per annum is consid-
ered low growth, and up to 4% as average
growth (Committee of Transport Officials
2013). Fast-growing areas are associated
with traffic growth of between 6 and 8%
per annum, or even over 8% in exceptional
cases. Traffic growth is particular to a
development, existing or future congestion,
and the urban or rural nature of the road.
The impact of traffic volume on speed is
also well established according to the prin-
ciples of Traffic Flow Theory. As vehicle vol-
ume increases, drivers reduce their speed to
cope with the increased density that brings
with it additional complexity in the interac-
tions between vehicles (Garber & Hoel
2015). The relationship between speed and
volume is usually described for short time
increments, with speed and volume aggre-
gated in intervals of between 20 seconds and
5 minutes (Treiber & Kesting 2013).
The longer-term relationship between
volume and speed, for example the impact
of annual traffic growth on average annual
speed, is less well understood. A study on
major roads in South Africa found that
average annual speed decreases annually on
the majority of roads (Bester & Geldenhuys
2007). This speed decrease is likely due to
an increase in traffic volume over time, but
this influence was not quantified. There
is significant variation in the relationship
between volume and speed on different
roads, influenced by road geometry and
driver population. This variability compli-
cates the formulation of a universal speed
estimation model.

Bruwer MM. A self-calibrating model to estimate average speed from AADT.

10 J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #953, 9 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a2
 Average annual speed is important
when conducting an economic assessment
of a transport facility because it is used
to estimate annual user costs (Margiotta
et al 1999). An economic assessment
considers whether the monetised economic
benefits provided by a project will exceed
economic costs. Many costs considered in
the economic evaluation of road projects
are directly associated with vehicle speeds.
Road User Costs (RUCs) include the cost of
operating a vehicle (fuel and oil consump-
tion), travel time costs and emissions, with
each component corresponding to speed.
This article considers the relation-
ship between average annual speed and
Annual Average Daily Traffic (AADT) on
major roads, specifically with two or more
lanes per direction and a speed limit of
120 km/h. The variability in the AADT–
speed relationship is addressed by the
formulation of self-calibrating speed pre-
diction models developed from empirical
data collected across South Africa, which
are therefore applicable to a wide range of
higher-order, high-speed roads. The speeds
of heavy vehicles and light vehicles can be
estimated separately, as well as the average
speed of all vehicles simultaneously.
the Application of
average speed IN
ECONOMIC EVALUATION
Transport economics assesses trans-
portation projects in monetary terms
to determine the economic benefit and
cost to establish if the project should
be implemented. The cost of transport
infrastructure can be grouped into three
subdivisions, namely the economic cost
of infrastructure, user cost and external
cost (Cape Metropolitan Transport 1994).
Infrastructure costs are directly related to
the transport facility, including construc-
tion and maintenance costs. User costs
(RUCs for road projects) include all costs
directly borne by road users, including
Vehicle Operating Costs and Time Costs.
External costs are not directly paid for
by users or developers of the facility and
include the cost of accidents, the social cost
of traffic congestion, such as time away
from family and environmental degrada-
tion (Quinet & Vickerman 2004).
Vehicle Operating Costs are a function
of speed because speed affects fuel and oil
consumption. Average speeds are used in
economic evaluation because average oper-
ating costs for all vehicles are estimated
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
(Thoresen & Roper 1996). Some literature
indicates that a single average speed for
all vehicle classes should be used (Heggie
1972), while other sources determine speed
and Vehicle Operating Costs for each vehi-
cle class separately (Pienaar & Bester 2008).
Factors affecting speed and existing speed
prediction models are discussed below to
provide background to this article.
Factors affecting average speed
Four categories of factors affect vehicular
speed – the driver, the vehicle, the roadway
and the environment (Winfrey 1969).
Drivers are influenced by driver age and
alertness, while vehicle factors such as
braking and cornering ability influence
speed. The roadway and environment
influence average speed through features
such as horizontal and vertical alignment
(affected by terrain), lane width, number
of lanes, shoulder width, speed limit and
pavement design.
Roadway features are used by the
Highway Capacity Manual (Transportation
Research Board 2000) to estimate Free Flow
Speed, i.e. the speed selected by drivers in
low-volume conditions when unhindered by
other vehicles. Free Flow Speed is adjusted
according to traffic volume to determine the
average speed at particular flow rates.
Factors that influence speed, such as
lane width, road alignment and speed limit,
are unlikely to change over time, unless
through infrastructure improvements.
However, RUCs are required for each year
of the economic analysis period (Pienaar &
Bester 2008). Traffic volume is a dynamic
factor that influences speed and is expected
to vary annually. Traffic volume varies
reliably with time and can be estimated
in the future by well-established traffic
demand forecasting models such as the
Four-Step Model in urban areas (Institute of
Transportation Engineers 2009) and traffic
growth rates calculated from historical data
for rural areas. Forecasted traffic volumes
should therefore be used to estimate average
speed, allowing RUCs to be calculated for all
years of the economic analysis period.
Existing speed prediction models
Various international speed prediction
models use AADT as an estimator variable
to determine average speeds (Margiotta
et al 1999; Lomax et al 1997; Schrank &
Lomax 2009).
A linear model to estimate speed from
Average Daily Traffic (ADT) on freeways
was proposed for the National Cooperative
Highway Research Program (NCHRP)
in the USA (Lomax et al 1997). ADT is
determined over a period shorter than a
year, identifying it from AADT. The 1997
Lomax model (Equation 1) was based
on empirical data and described average
speed (mph) in terms of ADT per lane
(1 000 veh/day/lane) and access frequency
(freeway interchanges per mile). The speed
prediction model was determined with a
Coefficient of Determination (R2) of 0.50.
U = 9 1.4 – 2.0(ADT) –
2.85(Access Frequency) (1)
A later speed model, also developed in
the USA, applies a stepwise prediction
model to estimate average speed on
freeways according to ADT as the only
predictor variable (Shrank & Lomax 2009).
This model estimates average speeds for
peak and off-peak periods separately,
and was developed from empirical data
from freeways in 29 cities. The stepwise
model predicts speed according to five
congestion levels – uncongested (ADT <
15 000 veh/‌day/‌lane), medium congestion
(15 000 < ADT < 17 500 veh/‌day/‌lane),
heavy congestion (17 500 < ADT <
20 000 veh/day/lane), severe conges-
tion (20 000 < ADT < 25 000 veh/day/
lane) and extreme congestion (ADT >
25 000 veh/‌day/lane). The Schrank &
Lomax (2009) model predicts that speeds
along uncongested routes are unaffected by
ADT. Speeds then decrease with increas-
ing ADT up to the point where extreme
congestion is encountered, after which the
model assumes that speed will remain con-
stant at 35 mph in the peak, and 40 mph in
the off-peak period.
In South Africa, a speed prediction
model that applies AADT was empirically
derived for the N17, a two-way, two-lane
road (Van As 2005). An adjustment factor
fc, presented by Equation 2, is multiplied
with Free Flow Speed to reduce speed
according to congestion. The factor also
takes road width into account through
coefficient a, defined as 80 080 or 154 570
for “lane plus shoulder width” of 3.7 m or
6.2 m respectively.
fc = e–AADT⁄a(2)
A method was developed by the South
African National Roads Agency Limited
(SANRAL) that makes use of a congestion
factor to adjust RUC directly based on the
ratio of AADT to daily capacity for various
11
 140

120

100

y = –0.0005x + 106.16
Average speed (km/h)

R2 = 0.1479
80

60

40

20

0
0 5 000 10 000 15 000 20 000 25 000
AADT per lane (veh/day/lane)

Figure 1 R
 elationship between Average Speed and AADT/ln “prediction” data (1994–2006)

road classes (Pienaar & Bester 2008). The congestion factor is multiplied with Free (ADT). ADT is a daily traffic volume deter-
stepwise function, presented in Equation 3, Flow Speed to estimate a lower average mined over a short time period such as a
calculates a congestion factor fc, which is speed. This assumes that the decrease in week or a month, rather than a full year
multiplied with RUCs and Time Costs to speed is of the same order as the increase as for AADT. Average annual vehicular
increase economic cost according to con- in both time costs and Vehicle Operating speeds are also recorded for all vehicles,
gestion levels (Pienaar & Bester 2008). Cost, which may not be accurate and could and for light and heavy vehicles separately.
be investigated in future research. All data entries not relevant to the
v study were removed, including data from
≤ 0.4: fc = 1.00
c roads with fewer than two lanes per direc-
Methodology
⎫v ⎫
tion and speed limits lower than 120 km/h.

0.4 ≤ ≤ 1.0: fc = ⎪ + 0.6⎪

1.15

⎭c ⎭
v This article empirically quantifies the Additionally, roads with an inadequate
(3)
c relationship between average annual speed representation of traffic data over a full
and AADT for multi-lane urban and rural year were excluded. The reported meas-
v freeways with a speed limit of 120 km/h ured ADT was assumed to be a satisfactory
> 1.0: fc = 1.61.15 = 1.72
c across South Africa. Furthermore, generic representation of AADT if 60% of the year
models are empirically formulated to was observed.
In the function described in Equation 3, predict future speeds on higher-order roads The data from CTOSs within close
v is AADT (veh/day) and c is a daily c­ apa­city for use in economic evaluation. proximity of toll plazas and freeway
estimate attributed to a particular road interchange ramps was also excluded from
class. Daily capacities are suggested to be Empirical data the dataset. A final set of 776 annual traf-
10 342 veh/day/lane for rolling terrain and The traffic information used in this study fic surveys from 153 CTOSs from across
13 575 veh/day/lane for flat terrain (Pienaar was obtained from the Comprehensive South Africa was available to calibrate
& Bester 2008). No adjustment is made to Traffic Observation (CTO) database, the speed prediction models (“prediction”
user and time costs where AADT is less collected by Mikros Traffic Monitoring dataset). The models were tested using data
than 40% of the daily capacity, correspond- (Pty) Ltd between 1994 and 2006 at CTO from 78 CTOSs around South Africa. Some
ing to AADT of around 5 500 veh/‌day/lane. Stations (CTOSs) on major roads across of these “test” CTOSs were included in the
It has been suggested that conges- South Africa. CTO data was also obtained “prediction” set (albeit at different years),
tion factors such as those described by from various CTOSs in 2008 and 2015 to while 45% were newly added.
Equation 3, which were formulated to test the accuracy of the proposed speed
be applied directly to RUC, can also be prediction models. Information provided Data analysis
applied directly to Free Flow Speed to esti- at each CTOS includes the station location, Linear regression analysis and assessment
mate average annual speed (Van As 2005). number of lanes, speed limit, percentage of of statistical significance using t-Test
In order to adjust speed, an inverse of the heavy vehicles and Average Daily Traffic Hypothesis Testing were the primary tools

12 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering

N1: Between Buccleuch and Allandale Interchanges
110
105
Average speed (km/h)
100
95
90
85
22 000 23 000
AADT per lane (veh/day/lane)
Figure 2 R
 elationship between Average Speed and AADT/ln at CTOS 565 along N1
N12: West of Gillooly's Interchange
94
93
Average speed (km/h)
92
91
90
89
88
11 000 11 500
AADT per lane (veh/day/lane)
Figure 3 R
 elationship between Average Speed and AADT/ln at CTOS 581 along N12
applied to empirical data to evaluate the
relationship between speed and AADT.
A t-Statistic with an absolute value of
greater than 1.960 indicates a statistically
significant relationship at a 95% confidence
level (Montgomery & Runger 2007). The
Coefficient of Determination (R2) was cal-
culated for regression models to determine
the applicability of the regression function
to describe the variability of the data.
Microsoft Excel was used for data analysis
and statistical investigation.
The capacity of the speed prediction
models to estimate future speed according to
traffic volume was tested using two quality
measures of error, namely Average Absolute
Speed Error (AASE) and Speed Error Bias
(SEB). These measures have been used exten-
sively to compare speed results (Gwara 2017).
Allowable error range for speeds compared
with the AASE and SEB error measures are
10 km/h and ±7.5 km/h respectively.
Analysis and speed prediction models
for average annual speeds of all vehicles are
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
y = –0.0038x + 189.59
R2 = 0.9621
24 000 25 000 26 000 27 000
y = –0.0024x + 120.52
R2 = 0.9708
12 000 12 500 13 000
presented in detail in this article. Similar
analysis methods were used to investigate
the average speeds of heavy vehicles and
light vehicles separately, and the main find-
ings and speed prediction models for heavy
and light vehicle speeds are summarised.
Empirical volume-
speed relationships
Correlation between AADT and
average annual speed
The relationship between AADT per lane
(AADT/ln) and average speed observed on
South African major roads is presented in
Figure 1 for the “prediction” study period
between 1994 and 2006. Average annual
operating speeds range from 80 to 120 km/h.
A maximum AADT/ln of 26 413 veh/‌day/‌lane
was recorded in 2006 on the Ben Schoeman
Freeway (N1) in Gauteng.
Figure 1 reveals the expected negative
correlation between average annual speed
and AADT/ln, i.e. average speed decreases
as AADT/ln increases. This agrees with the
finding of Bester and Geldenhuys (2007)
that average annual speed decreases annu-
ally. A linear regression analysis indicated
that the average annual speed of all vehicles
is significantly correlated to AADT/ln,
with a t-statistic of 11.590. The R2 value is,
however, low at 0.148, indicating that 14.8%
of the variability expressed in the dataset
is accounted for by the regression model.
The significant scatter of data points and
low R2 value results from speed-influencing
characteristics, such as lane width and
road alignment, that are not controlled or
accounted for in the dataset. Clearly, this
regression model cannot be used directly
to estimate speed from AADT/ln, as the
predicted speeds would be too inaccurate.
An interesting observation can be made
if Figure 1 is carefully considered, particu-
larly that a number of distinct datasets are
visible, forming discrete contours with rela-
tively constant gradients. These discrete
sets of data represent traffic information
at particular CTOSs over a few years. A
particularly visible dataset is isolated on
the far right in Figure 1, measured at CTOS
565 on the Ben Schoeman Freeway (N1) in
Gauteng between 1999 and 2006.
The relationship between AADT and
average speed at two CTOSs (CTOS 565
on the N1 at Kyalami, and CTOS 581 along
the N12 west of Gillooly’s Interchange) are
presented in Figures 2 and 3. Linear regres-
sion at these CTOSs yield R2 values of 0.96
and 0.97 respectively.
Two problems are associated with these
discrete datasets. Firstly, the regression
equations indicate a large discrepancy in
both the intercepts and the gradients, as
can be compared between Figures 2 and 3.
The second problem is related to the statis-
tical integrity of the regression model for
the entire dataset. In a regression analysis,
each predictor variable must be independ-
ent of other predictor variables. This is
not true for the regression model of the
entire dataset, because the data points from
distinct CTOSs are related to each other.
This problem is consequently referred to as
the “time series error” because it concerns
annual information in a time series for
each data group. Two options are avail-
able to eliminate the time series error –
firstly, to consider the relationship between
AADT/‌ln and speed at discrete CTOSs
separately over the years of analysis, and
secondly, to consider data for each year
separately at all CTOSs.
13
 1.5
1.0
0.5
Annual rate of change in speed (%)
–0.5
–1.0
–1.5
–2.0
–2.5
0 5 000
Figure 4 A
 nnual rate of change in speed compared to AADT/ln
Analysis of the correlation
between AADT/ln and average
annual speed at discrete CTOSs
The first option to eliminate the time
series error is to consider the relationship
between speed and AADT/ln over time
at each CTOS individually. The annual
change in AADT/ln and speed at discrete
stations is considered first, and then the
results of independent regression analyses
at discrete CTOSs are presented.
Annual traffic growth and change
in speed at discrete CTOSs
Annual traffic growth rate and change in
speed per annum were found to differ sig-
nificantly at individual CTOSs. The annual
change in AADT/ln and speed was analysed
at all CTOSs with four or more years of data.
For the “prediction” dataset, the annual
average traffic growth was observed to be
3.85%, in line with average traffic growth
in South Africa of 3 to 4% (Committee of
Transport Officials 2013). The minimum
traffic growth observed was –0.09% per
annum (N1 at the Mantsole CTOS between
1997 and 2001). A maximum growth rate
of 9.2% was observed on the N2 close to
the Umdloti Interchange between 2001 and
2006. The rate at which traffic increases per
year was compared to AADT/ln, but no cor-
relation was found, indicating that annual
traffic growth rate is not influenced by traf-
fic volume at the CTOSs investigated.
14
10 000 15 000
AADT per lane (veh/day/lane)
The average annual speed of all vehicles
was observed to decrease at an average
rate of –0.42% per annum. The minimum
annual change in speed was –2.32% at
the Kyalami CTOS on the Ben Schoeman
Freeway (N1) between 1999 and 2006
(incidentally also the CTOS with the high-
est AADT). The maximum speed change
was +1.27% per year (an increase in speed)
between 1999 and 2006 measured along
the N3 close to the Tugela Toll Plaza. The
AADT/ln at this location was relatively low
at 2 300 veh/day/lane in 2006.
There is a statistically significant cor-
relation between the annual rate of change
in average annual speed and AADT/ln
(t statistic –10.44, R2 0.51), as indicated in
Figure 4. The most recent year’s AADT/ln
at each CTOS was used in the comparison.
As can been seen in Figure 4, the annual
rate of change in speed increases at higher
AADT/ln. Speeds are therefore more
significantly influenced by traffic volume at
greater traffic levels.
From regression analysis of the relation-
ship presented in Figure 4, it was found that
the annual rate of change in average annual
speed for all vehicles was –0.066% per
1 000 veh/day/lane. At low traffic volumes
(less than 5 000 veh/day/lane), average
annual speed is not significantly influenced
by traffic volume, and speeds are equally
likely to increase as they are to decrease at
lower AADT/ln, as is evident in Figure 4. As
Volume 63 Number 2 June 2021
20 000 25 000
AADT/ln increases, the influence of volume
on speed becomes more pronounced. This
finding is in line with the SANRAL model
to estimate the impact of daily volume on
RUC, where no adjustment is suggested for
low volumes (Pienaar & Bester 2008).
AADT and average annual speed
relationship at discrete CTOSs
According to the first option to eliminate
the time series error, independent regres-
sion analyses were conducted at each
CTOS with five or more years of data
(96 individual stations).
The majority of CTOSs displayed a
significant negative correlation between
average speed and AADT/ln. In general,
the regression analysis at discrete CTOSs
resulted in high R2 values, attributed to the
fact that variables such as road alignment
and lane width remain constant at each
CTOS and therefore do not influence speed
variability. Of the 96 CTOSs analysed, 58%
produced an R2 value of greater than 0.6.
The majority of the CTOSs that presented
a poor correlation between AADT/ln and
average speed (R2 value lower than 0.6) had
a low AADT/ln of below 5 000 veh/‌day/‌lane.
At some CTOSs, there was an increase
in speed with increased AADT/ln, but all
CTOSs displaying this positive correlation
had a low AADT/ln (< 7 000 veh/day/lane).
This links with the findings of Figure 4
that speeds may increase annually at low
Journal of the South African Institution of Civil Engineering
Table 1 Annual Progression of the Speed vs AADT/ln Regression Models also had low AADT/ln and therefore higher
Year Intercept Gradient t-Statistic R2
speeds. Where heavy vehicle percentage
exceeded 15%, AADT/‌ln was consistently
2001 104.618 –0.000301 2.441 0.059 lower than 5 000 veh/‌day/‌lane.
2002 104.913 –0.000395 3.221 0.099 The speed of light vehicles and heavy
vehicles was compared to AADT/ln sepa-
2003 105.175 –0.000461 3.965 0.145
rately. The average speed of heavy vehicles
2004 104.581 –0.000536 5.029 0.190 was found to be uncorrelated to AADT/
2005 104.467 –0.000516 4.696 0.175 ln. The average speed of heavy vehicles
on all roads was maintained at an average
2006 106.017 –0.000693 6.085 0.322
of 80 km/h irrespective of traffic volume
(heavy vehicle speeds ranged between 70 and
90 km/h on individual roads). This is likely
0 due to additional speed limits imposed on
heavy vehicles in South Africa of 80 km/h.
–0.0001
The speed of light vehicles was found to
Gradient of AADT/ln–speed

–0.0002 be significantly correlated to AADT/ln and

was on average 2.7 km/h higher than the
–0.0003
speed of all vehicles (including heavy vehi-
–0.0004 y = –7E-05x + 0.1368 cles). The correlation between light vehicle
R2 = 0.9196 speed and AADT/ln is significant and is
–0.0005
slightly steeper than the regression model
–0.0006 for all vehicles. It was determined that the
light vehicle speed regression model fits
–0.0007 actual data better than for the speed of all
–0.0008
vehicles, indicated by its higher R2 value of
2000 2001 2002 2003 2004 2005 2006 2007 0.264 (0.148 for all vehicles).
Year of analysis

Figure 5 A
 nnual progression of the AADT/ln–speed relationship gradient
AADT/‌ln because traffic density is not high
enough to influence speed.
Annual progression of the
AADT/ln and average annual
speed relationship
This section considers the second method
to eliminate the time series error by using
only one data point per CTOS, but retaining
time series information. Individual regres-
sion models were produced for each year
from 2001 to 2006, using one data point
from each CTOS corresponding to that year.
This analysis does not consider traffic infor-
mation from 1994 to 2000, because of the
small sample of data available during these
periods resulting in insignificant relation-
ships. The results of the annual regression
analyses are summarised in Table 1. The
relationships are all significant according to
the t-statistic while the R2 values are low.
By considering traffic data for each
year separately, the annual progression of
the AADT/ln–speed relationship can be
investigated. Table 1 reveals that, while the
intercepts remain relatively constant (within
1.5 km/h), the gradients tend to become
steeper. The intercept is representative of Free
Flow Speed at low AADT/ln and is therefore
expected to remain constant over time as
there is no influence from traffic volume.
The annual progression of the gradi-
ent of the AADT/ln–speed relationship is
graphically indicated in Figure 5. There is
clearly a strong association between the
annual gradients of the AADT/ln–speed
relationship, with an R2 value of 0.92.
Impact of heavy vehicles
on speed progression
The CTO data includes Average Daily
Truck Traffic (heavy vehicles per day), the
percentage of heavy vehicles in the traf-
fic stream, as well as the average annual
speed of light vehicles and heavy vehicles
separately, allowing the impact of heavy
vehicles in the traffic to be evaluated.
Heavy vehicle population is described as
a percentage of total vehicles. The minimum
heavy vehicle percentage, 1.4%, was found
on the N12 in Gauteng west of Gillooley’s
Interchange, and the maximum of 40.3%
was observed on the N1 close to the Gariep
Dam. An increase in heavy vehicle percent-
age was curiously found to be associated
with an increase in average speed of all vehi-
cles. It was determined, however, that roads
with high percentages of heavy vehicles
Speed prediction modelS
The derivation of two speed prediction
models that estimate future average speed
according to predicted AADT/ln values
on higher-order roads in South African
is described in this section. The first
model uses findings from the investiga-
tion of the correlation between AADT/ln
and the average annual speed at discrete
CTOSs. The second model makes use of
the findings of the annual progression of
the AADT/ln and average annual speed
relationship. Models to estimate the aver-
age speed of all vehicles (identified by
subscript A) have been derived in detail in
this article. Models to estimate the average
speed of light vehicles (subscript L) are
derived through a similar process and the
resulting equations are presented at the end
of this section.
Model from correlations
at discrete CTOSs
The first speed prediction model consid-
ers the results of the evaluation of the
relationship between AADT/ln and aver-
age annual speed at discrete CTOSs. The
analysis identified that the rate of change
in speed per year is correlated significantly
to AADT/ln, according to Figure 3. The
annual rate of change of average annual

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 15
 120
115
110
Average speed (km/h)
105
100
95
90
85
80
0 5 000
Figure 6 G
 raphical explanation of the prediction model for all vehicle speeds
speed of all vehicles was observed to be
–0.066% per 1 000 veh/day/lane.
The relationship between annual change
in speed and AADT/ln informed the for-
mulation of an adjustment factor which can
estimate future speed for a particular road.
The adjustment factor is multiplied with the
speed of all vehicles observed in a base year
(UAb) to estimate the speed of all vehicles n
years into the future (UAf ). The adjustment
factor reduces base year speed according to a
future AADT/ln estimate (AADTf ) described
in Equation 4. Equation 4 is further referred
to as “Speed Prediction Model 1”.
⎫ –0.066 AADTf ⎫
UAf = UAb 1 + ⎪ ⎪ (4)
n
⎭ 100 1 000 ⎭
×
Model of annual progression
of the AADT/ln and average
annual speed relationship
The second speed prediction model is based
on the annual progression of the statisti-
cally significant liner relationship observed
between AADT/ln and speed. This rela-
tionship is described by Equation 5, which
relates the average speed of all vehicles, UAb,
at a particular base year b, to AADTb (which
is input as AADT/ln) of year b using an
intercept cb (Free Flow Speed) and a gradi-
ent mb. Equation 6 describes speed at some
time in the future, denoted by subscript f.
UAb = cAb + mAb . AADT b(5)
16
10 000 15 000
AADT per lane (veh/day/lane)
UAf = cAf + mAf . AADTf(6)
Investigation of the annual progression of
the AADT/ln and average annual speed
relationship identified that the intercepts
remain relatively constant (refer to Table 1).
It is therefore assumed that the intercepts
describing base and future predicted
speeds are equal, as per Equation 7.
cAb = cAf (7)
Further investigation showed that the gra-
dient of the AADT/ln and average annual
speed relationship decreases at a constant
rate of –7.3877 × 10 –5 per year. The gradi-
ent of the relationship for some future
year f can therefore be determined by
Equation 8 with n the number of calendar
years between b and f.
mAf = mAb – (7.388 × 10 –5)n(8)
Equation 9 results from combining
Equations 6, 7 and 8 to produce a model
that estimates the average speed of all vehi-
cles in year f, using base year and future
traffic characteristics.
UAf = U
 Ab – mAb . AADT b + [mAb –
(7.388 × 10 –5)n] . AADTf(9)
It would be inconvenient to have to
determine a specific gradient mAb for the
particular base year in every analysis. In
Volume 63 Number 2 June 2021
20 000 25 000
order to avoid having to define a gradient
for each base year, the gradient (–0.00051)
that describes the relationship between the
average speed of all vehicles and AADT/
ln (Figure 1), is used, resulting in Equation
10, which is further referred to as “Speed
Prediction Model 2”.
UAf = U
 Ab + 0.00051 . AADT b – [0.00051
+ (7.388 × 10 –5)n] . AADTf(10)
The first two terms of Equation 10
(UAb + 0.00051 . AADTb) estimate a unique
intercept for each road. The third term
estimates a unique gradient depending on
the number of years n. This procedure is
graphically explained by Figure 6. The red
points describe traffic on the N12 West of
Gillooley’s Interchange in Gauteng between
2000 and 2006, and the purple points on
the N1 at Rand Show between 2001 and
2006. The solid red and purple lines indicate
how unique intercepts are estimated from
2001 traffic data. The gradient of these solid
lines is the same as the general regression
equation (solid grey line). The second step
determines a gradient based on the five
years between 2001 and 2006, as indicated
by the dashed lines, to estimate 2006 speed
from a predicted AADT/ln value.
Speed prediction model 2 is self-­
calibrating. A particular intercept is not
stipulated and so a unique model can be
produced for every major road in South
Africa. The model is calibrated for a
Journal of the South African Institution of Civil Engineering
particular road by using actual speed and
AADT/ln values, providing information
about the road’s unique volume–speed
relationship. This allows the prediction
model to implicitly take roadway char-
acteristics into account, such as terrain,
lane and shoulder width, the number of
interchanges and alignment.
Speed prediction models for
heavy and light vehicles
The speeds of heavy vehicles (denoted by
subscript H) are not affected by traffic vol-
ume. Heavy vehicle speeds observed during
the base year can therefore be maintained
throughout the analysis period according
to Equation 11:
UHb = UHf(11)
The annual rate of change of average
annual speed of light vehicles was observed
to be –0.071% per 1 000 veh/day/lane from
evaluation of the relationship between
AADT/ln and average annual speed of
light vehicles at discrete CTOSs (t statistic
–10.46, R2 0.51). Similar to Equation 4 for
all vehicle speeds, Equation 12 estimates
the average speed of light vehicles accord-
ing to “Speed Prediction Model 1”.
⎫ –0.071 AADTf ⎫
ULf = ULb 1 + ⎪ ⎪ (12)
n
⎭ 100 1 000 ⎭
×
The annual progression of the relationship
between AADT/ln and the average speed
of light vehicles was analysed similarly to
the speed of all vehicles. The gradient of
this relationship decreases at a constant
rate of –7.3298 × 10 –5 per year. Similar to
Equation 10 for all vehicles, Equation 13
estimates light vehicle average speed
according to “Speed Prediction Model 2”.
U Af = U
 Ab + 0.00077 . AADT b – [0.00077
+ (7.330 × 10 –5)n] . AADTf(13)
Model testing
The two speed prediction models to
estimate future speed from AADT/ln were
evaluated using a “test” dataset collected
at 78 CTOSs around South Africa in 2008
and 2015. 2015 speeds were estimated from
2008 data, and then compared to actual
speed measurements of 2015. In addition
to the two models formulated for both all
and light vehicle speeds, comparison is also
made to speed estimates calculated using
the SANRAL method (Pienaar & Bester
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
Table 2 Q
 uality measures of speed prediction models
MODEL
Speed Prediction Model 1 (all vehicles)
Speed Prediction Model 1 (light vehicles)
Speed Prediction Model 2 (all vehicles)
Speed Prediction Model 2 (light vehicles)
Speed Prediction for heavy vehicles
SANRAL RUC adjustment factor
Allowable error
120
2015 estimated speeds (km/h)
110
100
90
80
70
70
Figure 7 C
 omparison of speeds estimated using Speed Prediction Model 1
2008). This method was formulated to be
applied on all higher-order roads in South
Africa directly to RUC and not speed; how-
ever, Van As (2005) suggested that similar
factors can be applied to speed directly.
Two quality measures of error, namely
Average Absolute Speed Error (AASE) and
Speed Error Bias (SEB) are used to evaluate
speeds estimated by the prediction models.
Allowable error range for speeds compared
with the AASE and SEB error measures are
10 km/h and ±7.5 km/h respectively. The
results of the error measures are presented
in Table 2.
Both Speed Prediction Model 1 and Speed
Prediction Model 2 (for all and light vehicles)
estimate speeds that are well within the error
allowances of both the AASE and SEB meas-
ures. Speed Prediction Model 1 performs
slightly better than Speed Prediction Model 2
in both instances. The speed estimation for
all vehicle speeds is slightly more accurate
than speed estimation for light vehicles
only. The error terms for the speed of heavy
vehicles estimated according to Equation 11
are also presented in Table 2. The AASE and
SEB for heavy vehicle prediction are very low,
AASE (km/h) SEB (km/h)
3.463 –0.480
5.094 –2.216
4.028 –1.340
5.652 –3.003
2.574 0.298
11.6989 –7.61942
10 ±7.5
R2 = 0.723
80 90 100 110 120
2015 measured speeds (km/h)
indicating that the assumption that speed of
heavy vehicles remains constant irrespective
of AADT/ln is valid. Estimates of speeds cal-
culated from the SANRAL RUC adjustment
factor do not adequately represent actual
speeds; however, it must be remembered that
this model was not formulated to predict
future speeds, rather the increase in RUC.
Comparison of 2015 actual speeds to
2015 estimated speeds for all vehicles is
presented in Figures 7 and 8 for Speed
Prediction Model 1 and Speed Prediction
Model 2, respectively. For Model 1, an R2
value of 0.723 is determined, indicating a
relatively good relationship between esti-
mated speeds and those actually measured
in 2015. Model 2 resulted in a slightly lower
R2 value of 0.698, indicating marginally
more variability between estimated and
actual speeds. Considering the very high
variation in observed speeds on higher-
order roads across South Africa (refer to
Figure 1), these models both do well in
estimating future speeds. The red diagonal
lines in Figure 7 and 8 indicate a gradient
of 1.0. The regression lines of both Speed
Prediction Model 1 and Speed Prediction
17

120
2015 estimated speeds (km/h)
110
100
90
80
70
70
Figure 8 C
 omparison of speeds estimated using Speed Prediction Model 2
Model 2 are slightly less steep than the
diagonal, indicating that both prediction
models slightly overestimate speeds lower
than 95 km/h and slightly underestimate
speeds higher than 95 km/h.
Conclusions and
recommendations
The objectives of this article were firstly to
quantify the relationship between AADT
and average annual speed on major highways,
and secondly, to generate prediction models
to forecast speed from AADT, applicable on
a range of higher-order roads for use in eco-
nomic analysis of road infrastructure.
A statistically significant relationship
between AADT/ln and average annual
speed was identified on major highways
across South Africa. However, the relation-
ship displayed a high level of variability and
a time series error in which distinct groups
of predictor variables (AADT/ln) were
related. To eliminate this error, data was
firstly considered at discrete positions on
the road network, and secondly, the annual
progression of the relationship between
AADT/ln and speed was evaluated. At
discrete locations, the AADT/ln–speed
relationships were significantly less vari-
able; however, each location presented a
unique correlation, influenced by road
characteristics, complicating the formula-
tion of a universal speed prediction model.
The annual progression of the AADT/ln–
speed relationship revealed that, while Free
Flow Speed remained relatively constant
year after year, the rate at which speed is
18
R2 = 0.6983
80 90 100 110 120
2015 measured speeds (km/h)
affected by AADT/ln increases annually at
a constant rate.
Two novel speed prediction models
were proposed in this article. Both models
require the input of AADT/ln of the
forecasted year to estimate a future aver-
age annual speed. Both models include a
calibration variable, i.e. the average annual
speed in the base year of analysis. The
calibration step accounts for the variation
in the AADT/ln–speed relationship result-
ing from specific roadway features. This
calibration is unique to speed prediction
models and increases the reliability of these
models to estimate speed considerably.
Separate prediction models were developed
to estimate the average annual speed of
all vehicles and light vehicles. The speed
of heavy vehicles was found to remain
constant throughout the analysis period,
irrespective of AADT/ln.
The proposed speed prediction models
can be used to estimate average annual speed
for all vehicles and light vehicles from fore-
casted AADT/ln values for use in economic
assessment. These models are applicable to
all higher-order roads in South Africa with
two or more lanes per direction and a speed
limit of 120 km/h. It is recommended that
further research be conducted to explore the
applicability of these prediction models in
other countries, and to calibrate the models
for other types of roads.
REFERENCES
Bester, C & Geldenhuys, F 2007. Speed trends on major
roads in South Africa. Proceedings, 26th Southern
Volume 63 Number 2 June 2021
African Transport Conference 2007, 9–12 July 2007,
Pretoria, pp 600–609.
Cape Metropolitan Transport 1994. Cape Metropolitan
Guidelines for Conducting the Economic Evaluation
of Urban Transport Projects. Cape Town:
Municipality of Cape Town.
Committee of Transport Officials 2013. TMH 17:
South Africa Trip Data Manual. Pretoria: SANRAL.
Garber, N & Hoel, L 2015. Traffic and Highway
Engineering, 5th ed. Stamford, CA: Cengage
Learning.
Gwara, A 2017. Validation of TomTom historical average
speeds on freeway segments in Gauteng, South Africa.
MEng Dissertation. University of Stellenbosch.
Heggie, I 1972. Transport Engineering Economics.
London: McGraw Hill.
Institute of Transportation Engineers 2009.
Transportation Planning Handbook, 3rd ed.
Washington, DC: Institute of Transportation
Engineers.
Lomax, T, Turner, S & Shunk, G 1997. Quantifying
Congestion, Vol 1, Final Report. National Cooperative
Highway Research Program (NCHRP) Report 398,
Washington, DC: Transportation Research Board.
Margiotta, R, Cohen, H & DeCorla-Souza, P 1999.
Speed and delay prediction models for planning
applications. Proceedings, 6th National Conference
on Transportation Planning for Small and Medium-
Sized Communities, Spokane, WA.
Montgomery, D & Runger, G 2007. Applied Statistics
and Probability for Engineers, 4th ed. Hoboken, N J:
Wiley.
Pienaar, W & Bester, C 2008. Specialist socio-economic
impact study of the proposed N2 Wild Coast Toll
Highway. Report prepared for CCA Environmental,
on behalf of SANRAL by the University of
Stellenbosch.
Quinet, E & Vickerman, R 2004. Principles of
Transport Economics. Cheltenham, UK: Edward
Elgar Publishing.
Schrank, D & Lomax, T 2009. 2009 Urban Mobility
Report. College Station, TX: Texas Transport
Institute.
Thoresen, T & Roper, R 1996. Review and enhancement
of vehicle operating cost models: assessment of non
urban evaluation models. Research Report ARR 279.
Victoria, Australia: Australian Road Research
Board.
Transportation Research Board (TRB) 2000. Highway
Capacity Manual. Washington, DC: TRB.
Treiber, M & Kesting, A 2013. Traffic Flow Dynamics:
Data, Models and Simulation. Heidelberg,
Germany: Springer.
Van As, C 2005. Travel Time Estimation on the N17.
Pretoria: SANRAL.
Winfrey, R 1969. Economic Analysis for Highways.
Scranton, PA: International Textbook Company.
Journal of the South African Institution of Civil Engineering
 TECHNICAL PAPER
Static pile capacity from
the rapid test via the
Journal of the South African
Institution of Civil Engineering

unloading-point method
ISSN 1021-2019
Vol 63 No 2, June 2021, Pages 19–26, Paper 1120

DR IRVIN LUKER (Pr Eng, MSAICE), who is a

I Luker Senior Lecturer at the University of the
Witwatersrand in Johannesburg, joined the
university’s academic staff after having
spent nine years in consulting and
Laboratory measurements were done of the end-bearing and side-shear stresses on small piles contracting, and then in 1986 graduated
with a PhD. Over the years he has lectured
subjected to “static” and rapid load tests. Interpretation, by the unloading-point method, of
and published on a variety of subjects, but
the results of rapid tests were done to estimate what the behaviour under a static test would mainly on geotechnical engineering. Since 2016 he has been working
have been if it had been done instead. Comparisons were made of the estimated static test full-time on research topics in the university’s civil engineering laboratory
behaviour to the actual static test behaviour to determine empirical values of a factor to adjust and is currently developing techniques and equipment to enable the rapid
testing of foundation piles to be done inexpensively in South Africa.
estimates closer to actual static test results. The dependence of the empirical factor on soil type,
the liquid limit of plastic soils and water content is shown. Contact details:
School of Civil and Environmental Engineering
University of the Witwatersrand
PO Wits
INTRODUCTION The measurements of force and settle- 2050
The four basic types of pile load test are ment in all four basic types of pile test need South Africa
T: +27 11 717 7114
commonly called: (i) maintained load to be interpreted to predict how the pile
E: irvin.luker@wits.ac.za
(or static), (ii) constant rate of penetra- will behave under its working load. Because
tion (CRP), (iii) rapid, and (iv) dynamic. of the wave effects in the dynamic test its
However, in the maintained load test, interpretation is difficult, requiring complex
increments of load are added when the computer software, knowledge of empirical
rate of penetration has reduced to (typi- correlations and an experienced interpreter.
cally) 0.25 mm/hour. Therefore, like the In comparison, the unloading-point method
CRP, it is also a test in which the rate of interpretation of the rapid test is simpler.
of penetration is specified, although It can be concisely described as follows:
extremely slow. The rapid and dynamic Figure 1 shows, for a rapid test, a typical
tests are both done quickly, and up graph of total force measured at the head of
to twelve pile tests per day have been a pile versus displacement, where:
reported. The fundamental difference
between the rapid and dynamic types is Total force = s oil resistance under static
the length of time over which the testing conditions + extra soil
force is applied. For the rapid test it lies in resistance from its rate of
the approximate range of 0.02 to 0.2 sec, straining + inertia force
(depending on pile length and material), from mass × acceleration. (1)
and for the dynamic test it is approximate-
ly 0.005 sec. In the dynamic test a force At the unloading point the velocity is zero,
pulse of short wavelength travels down the so the second term of the soil resistance
pile shaft (in compression) and back up (in in Equation 1 is zero. The acceleration is
tension). In the rapid test, the force pulse measured throughout the test and the mass
has a wavelength that is sufficiently long can be estimated, so the third term can be
for the whole pile to be in compression calculated. Subtraction of the third term
simultaneously and no reflected tension from the measured total force gives the soil
wave effects are seen. resistance under static conditions, at the
Further information about the four point of maximum settlement in the test.
types of test can be found in Fleming et al A formal statement of this procedure
(1994) and Byrne et al (2019). More specific is given in Hölscher et al (2012) and
information about the rapid test is given by ISO 22477-10:2016 (ISO 2016).
Middendorp et al (1992), Middendorp and Values of static resistance calculated
Bielefeld (1995), and Hölscher et al (2012). using this procedure, when compared to
Standards defining the method of execu- those from the maintained load (static) test,
tion of the rapid test are ISO22477-10:2016 have been found to be accurate for coarse,
(ISO 2016), ASTM D7383-08 (ASTM 2008) non-cohesive soil, but to overestimate the
and JGS 1815-2002 (JGS 2002). resistance of cohesive soils (Hölscher et al Keywords: piles, load test, clays, dynamics, empiricism

Luker I. Static pile capacity from the rapid test via the unloading-point method.
J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #1120, 8 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a3 19
2012; Weaver & Rollins 2010). Adjustment
of this overestimate is done by an empirical
factor μ, defined as follows:
Unloading point
force measured in a static test
μ= (2)
static force interpreted from a rapid test

Force at head of pile

In a review of comparisons between rapid
and static tests by Hölscher et al (2012) a
value of μ = 0.66 is given “for all clay”, with
a coefficient of variation of 49%. These
values are repeated in ISO 22477-10:2016
(ISO 2016). Brown and Powell (2013) stated
that more information about the variation
of μ with soil type was needed. They chose
liquid limit (LL) to represent the potential
for rate effects in cohesive soils, and from Pile displacement
published results of six sources of tests
where the LL was known they suggested: Figure 1 Typical result of a rapid test on a pile

μ = –0.0033 × LL + 0.69 (3)

Slowly applied force at displacement
which has a Pearson R2 correlation coef- 0.25 mm/hour or rapid test force
ficient of 0.6.

These coefficients (49% and 0.6) seemed to

the present author to be too high and too Accelerometers
low, respectively, for satisfactory applica-
tion in practice, so the work described in
this paper was done to try to improve the
accuracy with which μ for the unloading-
point method (UPM) can be estimated.
It is noted that two other methods of Displacement transducers
interpretation of the rapid test have been
proposed and have been reviewed by Brown
and Powell (2013). They applied all three
methods to maintained load and rapid field Steel guide tube
tests on two clay sites. Perhaps because of
limited data they did not give broad recom- Compacted soil
mendations for which method is preferable.
It is suggested by the present author that
0 101
the UPM has the benefit of simplicity. It is
PVC tube
also noted that it is the only method recom-
mended by ISO 22477-10:2016 (ISO 2016).

0 300

CHOICE OF METHOD OF
INVESTIGATION OF μ
The first choice made was between full-
scale site tests where rapid tests are com-
pared to maintained load tests, or smaller-
scale laboratory tests. To be of practical use,
values of μ for a wide range of soil types are
needed. Site tests are very expensive and the Stiff packing
types of soil around a pile can vary greatly
along its length, therefore laboratory testing
on a small diameter pile was chosen.
The second choice presented, if a whole
pile were to be tested, was the value of
the ratio of length to base diameter of the Figure 2 Apparatus for end-bearing tests

20 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 18
16
14
12
10.4 kN
End-bearing force (kN)
10
0
–2
0 5
Figure 3 End-bearing slow loading tests sequence
pile, λ. Influencing this choice were: (i) the
soil mechanics of side shear are different
from end bearing, (ii) the magnitude of
the influence of loading rate on μ for side
shear has been reported to be greater than
that on μ for end bearing (Brown 2004),
and (iii) a wide range of λ occurs in piling
practice, and a sufficient number of tests in
the present investigation could not be done
to represent that range. After considering
these influences it was decided to do
separate tests for side shear and end bearing.
The third choice presented was that of
the parameters to be measured of the soils
in the beds in which the pile was tested, for
possible use in empirical relationships to
predict μ. Influencing this choice were: (i) the
soil parameters should be easily obtained,
such as those that are usually measured in a
site investigation, (ii) the parameters likely to
have a significant influence on the value of μ
must be included, and (iii) Brown and Powell
(2013) had used liquid limit to characterise a
cohesive soil in their empirical equation for μ.
(The soil parameters chosen to be measured
in this study are given in Table 2 on page 24.)
APPARATUS AND PROCEDURE
FOR END-BEARING TESTS
Figure 2 gives details of the apparatus. The
closed end of the inner tube bears onto the
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
A
2.81 mm
10
Displacement (mm)
compacted soil below it. The outer tube
allows the top of the compacted soil to be
above the bearing level. Friction between the
tubes was reduced to a negligible amount by
soft grease. Bearing force on the inner tube
was measured by the same load-cell in both
very slow penetration tests (so-called “static”
tests) and rapid tests. Displacement of the
inner tube relative to the soil container was
measured by two linear potentiometers.
Two accelerometers were fixed diametrically
opposite on the inner tube to measure accel-
eration in the direction of the bearing force.
Soil was compacted by hand into the
PVC tube container by a rammer of 75 mm
diameter and mass 4.5 kg. The water content
of each soil was adjusted to the estimated
optimum for compaction, resulting in all
soil test beds being partially saturated and at
water content at or below their plastic limit.
The boundary of the soil below the bear-
ing level of the pile was at five times its diam-
eter, so that the mechanics of the soil were
not significantly influenced by this boundary.
However, the diameter of the container was
only three times the pile diameter, therefore
this boundary did have an influence on the
soil bed’s behaviour. This influence was con-
sidered acceptable because the purpose of the
testing was comparison of behaviour in two
types of test, and both types would have the
same boundary conditions.
15 20
The slowly applied bearing force was
produced by advancing the base of the pile
at a rate close to 0.25 mm/hour, using an
electric motor and a custom-built series
of gears ending in a slowly turning nut on
a threaded shaft. The rate of 0.25 mm/hr
was chosen because it is close to the rate
of settlement at which load increments are
added in standard maintained load (static)
tests, e.g. ASTM D7383-08 (ASTM 2008)
and SANS 1200F (SANS 1983).
The rapid test force was produced by
dropping a 100 kg mass of steel onto a
spring at the head of the pile. Because the
stiffness and strength of the different soil
beds varied, the height of drop of the mass
needed to deliver the desired force also var-
ied. This variation in drop height caused
variation in the velocity of the pile base,
which is discussed further on in this paper.
The time period over which a force
should be applied for a test on a pile to be
a rapid test has been defined (Middendorp
& Bielefeld 1995) as required to be at least
10 times the time needed for a stress wave
to travel the length of the pile. So, the ques-
tion arises: are the time periods over which
force was applied in the end-bearing tests
sufficiently close to those that would occur
in rapid tests on whole piles? The range of
the time periods in the end-bearing tests
was 0.02 to 0.025 sec. Considering piles
21
 made of good quality concrete with a wave
End-bearing force, adjusted for inertia forces (kN) 14 velocity of 4 000 m/s, the minimum pile
Unloading
point length to exactly comply with the rapid test
12 11.2 kN
definition is then given by:
10
4 000
L = 0.02 × = 8 m.
8 10

6
This length is realistic for a concrete pile,
4 therefore the time periods over which
end-bearing stress was applied to the soil in
2 these tests can be considered realistic for a
rapid test.
0
2.81 mm
–2
Sequence of test types
0 0.5 1.0 1.5 2.0 2.5 3.0 Ideally one would test the pile in a bed of a
Displacement (mm) particular soil using one test method, remove
the pile and re-establish it in an identical
Figure 4 End-bearing rapid loading test bed of soil, then test it using the other test
method. However, it was found to be impos-
sible to get sufficiently identical conditions of
Slowly applied force at displacement soil compaction in successive tests.
0.25 mm/hour or rapid test force The solution adopted is illustrated in
Figure 3 (page 21), showing measurements
from one of the tests. The pile was first
loaded slowly (0.25 mm/hr as in a static test)
PVC tube
to a displacement of 5.6 mm, unloaded, then
slowly reloaded again to a total displacement
of 12.7 mm and unloaded. A rapid reload
250

test was then done (for clarity, not shown

Steel tube in 3 lengths
in Figure 3). A second slow reload test was
then done. As can be seen in Figure 3, the
soil conditions encountered by each reload
Aluminium loadcell and test are different, because of the action on
50

accelerometers
the soil of the preceding test. To compare the
behaviour in the rapid test to that in the slow
0 90
test, the graph of slow test behaviour at dis-
placements starting from point A on Figure 3
Compacted soil
is assumed would rise up to the green line, in
300

Glued-on sand a similar manner to the slow reload tests on

either side, then follow that green line.
The force versus displacement behav-
iour of the rapid test that was actually
Aluminium loadcell done, starting from point A on Figure 3,
50

is shown in Figure 4. Note that the force

values on the ordinate axis have been
corrected for the contribution to the force
measured at the load-cell from acceleration
250

Displacement transducer and inertia of the inner tube and impact

plate on top of it. Displacement in both
types of test was measured by transducers,
Concrete block during soil but in the rapid test was also calculated
compaction replaced by
polystyrene foam with from integration of the measured accelera-
0 100 hole during tests tion. Values were very similar but offset
by a small time difference when plotted
on a graph of force versus time. Because
0 300 the values from integration of acceleration
give a smoother line than those from the
displacement transducers, they are used for
Figure 5 Apparatus for side-shear measurement all the rapid tests reported in this paper.

22 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 4
3
Shear force on pile's middle third (kN)
0
0 5
Initial loading
Figure 6 Force versus displacement results of a typical series of slow tests measuring side-shear stress
Consider further this example: at the
unloading point the displacement on Figure 4
is 2.81 mm and the force value (confirmed on
a graph of force versus velocity, at ­velocity = 0)
is 11.2 kN. On Figure 3 the slow test force
from the interpolated green line at a displace-
ment of 2.81 mm from point A is 10.4 kN.
Therefore, the value in this example of μ
(from Equation 2) is 10.4/11.2 = 0.93.
RESULTS OF END‑BEARING TESTS
Twelve different soils were used for tests of
side-shear and end-bearing behaviours (char-
acteristics of which are given in Table 2 on
page 24). Soil Numbers 1, 2 and 7 were used
for end-bearing tests, chosen because 2 and 7
are at the extremes of fineness and coarseness
of the twelve, and Soil No 1 is approximately
midway. The results are given in Table 1.
The range of variation of μ is small
compared to that reported by Brown and
Powell (2013) for whole piles, and the
mean is close to 1. This indicates: (i) the
unloading-point method is accurate for end
bearing, and (ii) variation in soil type has
little effect on the accuracy.
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
10
Displacement (mm)
1st slow reload 2nd slow reload 3rd slow reload
Table 1 Results of end-bearing tests
Soil USCS
μ concrete block at the base of the tube was
number classification
2 CH 0.88
1 SC 1.06
7 SP 0.93
APPARATUS AND PROCEDURE
FOR SIDE-SHEAR TESTS
Figure 5 gives details of the apparatus. The
pile was made from three lengths of steel
tube, between which were two pieces of
aluminium tube. The five tubes were strongly
glued together. Both aluminium tubes had
strain gauges on their internal faces so that
they functioned as load-cells to measure
longitudinal force, and the upper aluminium
tube also had two accelerometers for the
longitudinal direction. Soil was compacted
round the pile inside the same 300 mm
diameter tube used for the end-bearing tests,
against the reaction of a concrete block at the
bottom of the tube. Both pile and container
tube had rough surfaces of glued-on sand to
15 20
4th slow reload
prevent slip and ensure shear was within the
soil. Before beginning testing of the pile, the
replaced by a block of expanded polystyrene
foam with a central hole 3 mm bigger in
diameter than the bottom steel tube of the
pile. During all tests on the pile, its down-
ward displacement was freely into the hole
in the polystyrene foam whose purpose was
to prevent spalling of the bottom face of the
soil annulus. Although the shear stiffness of
the foam was different from the soil, because
the shear behaviour of the middle third of
the length of the pile was being examined,
300 mm away from the foam, the effect
of the foam on that shear behaviour was
assumed to be negligible.
The procedure adopted is similar to that
used for the end-bearing tests and is illus-
trated in Figure 6. The pile was first loaded
slowly (0.25 mm/hr as in a static test) to its
maximum capacity, unloaded, then slowly
reloaded again to its maximum capacity, and
unloaded. A rapid reload test was then done
(for clarity not shown in Figure 6). A second
slow reload test was then done. As can be
seen in Figure 6, the shapes of the graphs of
23
 8

6
Shear force on pile's middle third (kN)

–1
–0.2 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Displacement (mm)

Figure 7 Shear force versus displacement of a typical rapid reload test

the initial loading and the two slow reload In these tests, for one pile embedment test done between the first and second reload
tests are different. Consider the graphs of the in a soil, four slow reload tests were done, tests shown in Figure 6. The force has been
initial slow loading and the first slow reload with three rapid tests between them. From adjusted to remove inertia effects. Its shape
tests. They are different because the soil con- comparison of the behaviours in the two is similar to Figure 4 for an end-­bearing test,
ditions that the reload test encounters have types of test, three values of μ for that soil except that it is more flat-topped, as can be
been altered by the action of the initial load- were obtained. The mean of the three has expected from the flat-topped shape of the
ing. For the same reason, the graphs of the been taken as representative of the soil and slow tests results shown in Figure 6.
two slow reload tests are different from each its condition in those tests. Tests to compare side-shear behaviours
other. However, they are sufficiently similar in the slow and rapid tests were performed
for the assumption to be made that the mean in twelve different soils. Measured
of them represents the slow test behaviour of RESULTS OF SIDE-SHEAR TESTS properties of the soils and their density and
the soil that would have been seen if a slow Figure 7 shows a typical graph of force versus moisture content conditions during the
test had been done instead of the rapid test displacement for the side-shear resistance tests are shown in Table 2, together with
that was actually done. on the middle length of the pile in the rapid the values of μ.

Table 2 Soils used in the testing, measured characteristics and values of μ

USCS Plastic Liquid Grading % Dry density Water Degree of
Soil no µ
classification limit limit Gravel Sand Silt Clay kg/m3 content % saturation

1 SC 18 28 16 50 22 12 1542 9.9 0.37 0.79

2 CH 24 55 2 19 27 52 1083 17.0 0.31 0.64
3 SC 20 31 22 58 14 6 1632 5.6 0.24 0.81
4 CL 23 34 16 36 37 11 1569 18.4 0.71 0.79
5 SC 15 25 28 55 11 6 1435 4.4 0.14 0.80
6 CL 19 45 3 48 15 34 1395 19.5 0.57 0.81
7 SW n.p n.p 41 49 10 0 1932 2.8 0.20 1.05
8 SC 29 60 9 52 14 25 1169 26.7 0.56 0.61
9 CL 13 24 0 55 24 21 1671 11.1 0.50 0.65
10 SC-SM 14 21 7 71 14 8 1853 11.1 0.68 0.80
11 SC 13 29 4 53 15 28 1471 13.0 0.43 0.80
12 SP n.p n.p. 3 47 41 9 1550 13.0 0.50 0.83

24 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering

0.9
0.8
0.7
0.6
Values of Mu
0.5
0.4
0.3
0.2
0.1
0
0 20
Brown & Powell 2013 Optimum WC
Figure 8 μ versus liquid limit
The values of μ for the non-plastic
soils numbers 7 and 12 are close to 1, as
expected. However, the range of μ in the
ten plastic soils is from 0.61 to 0.81. Clearly
some means of choosing the appropriate
value of μ for a site of a rapid pile test is
required. In the absence of mathematical
theory representing the variation in soil
conditions, empirical correlation is invoked.
EMPIRICAL CORRELATION
BETWEEN μ FOR SIDE-SHEAR
AND SOIL PROPERTIES
Following the example of Brown and
Powell (2013) μ from these side-shear tests
was first plotted against liquid limit, as
shown in Figure 8, which also incorporates
their values for whole piles plotted as grey
diamond-shaped points. The red points are
present results of tests done at water con-
tent close to the optimums for compaction
of the soils.
It is clear that the trend of values of μ
in the red points is higher than those of
Brown and Powell (2013). It was thought
that this may have been caused by differ-
ent amounts of water in the soils, because
the field tests reported by them were done
on sites with high water tables, so it may
be assumed that the majority of the soil
affected by the piles was saturated.
To test this hypothesis, tests on soils
numbers 8 and 11 (whose liquid limits are
60 and 29 respectively), were done at three
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
correlations with WC and S. The correlation
of μ with the rate of shearing was also
investigated, using the parameter of the
maximum pile velocity in each test.
Clearly μ is best correlated with the
plasticity parameters, with liquid limit
(LL) being the better one. It is similarly
correlated with WC. Correlation with
aspects of the grading curve are not good,
and degree of saturation is, surprisingly,
poorly correlated with μ. Also surprising
is the small correlation of μ with the
velocity of the pile during the test, in this
case represented by the maximum velocity
value.
To get the most accurate empirical
equation for μ from the measured results,
various combinations of parameters were
investigated, but including more than LL
40 60 80 100 and WC produced negligible improvement.
Liquid limit Therefore a function of the form (Equation 4)
Soils 8 & 11 higher WC Calc'd Mu at sat'n was chosen:
μ = m1.LL + m2.WC + m3.LL.WC + m4 (4)
different degrees of saturation. In Figure 8 When the m1 to m4 factors were deter-
the blue points at liquid limit = 60 and 29 mined from the data of the red and blue
show the resulting decrease in μ towards points on Figure 8, the R2 value was 0.88.
the region of the tests done (it is assumed) This seems relatively good, but comparing
in saturated soil. the calculated values of μ to the actual
It is therefore concluded that in an values of μ showed that the function
empirical evaluation of μ for a soil, its water overestimates μ at high WC. This is
content must be one of the parameters because only two soils (numbers 8 and 11,
included. LLs 60 and 29 – blue points on Figure 8)
In the present study the linear correla- have test data at high WC.
tion of μ with a variety of parameters of soil To obtain a function for μ with broad
characteristics was investigated using least application in practice, the green points
squares regression analysis. Table 3 gives in Figure 8 were generated for the soils
the values of R2 seen for the measured soil at their saturation WC, using the linear
properties for the tests shown by the red correlation between WC and μ whose R2
points in Figure 8, for all parameters except value was 0.81 in Table 3. The factors of
water content (WC) and degree of satura- Equation 4 were then found using the data
tion (S). The tests shown by the blue points of all points from the present side-shear
were included with the red points for the tests on Figure 8, (i.e. all except Brown
and Powell’s data, the grey points), giving
Table 3 Parameters linearly correlated with μ Equation 5.
Soil parameter Pearson R2
μ=–
 1.755e-3.LL – 9.762e-3.WC +
D10 grain size 0.14 2.465e-5.LL.WC + 0.920 (5)
D50 grain size 0.19
One would normally show the line of an
% < 0.002 mm 0.43
empirical equation superimposed on a
% < 0.425 mm 0.28 graph of the data points from which it
Liquid limit 0.91 was derived, but this is impractical in the
present case where the equation has two
Plasticity index 0.54
independent variables, LL and WC. Instead,
Water content 0.81 the ability of Equation 5 to reproduce the
Degree of saturation 0.39 measured values of μ is shown in Figure 9,
which plots both actual and calculated
Max velocity of pile 0.15
values of μ against LL.
25

0.9
0.8
0.7
Mu values
0.6
0.5
0.4
0.3
10 20 30
Liquid limit
Measured data
Figure 9 Comparison of calculated and measured μ values
DISCUSSION
Because the grey points on Figure 8 are for
whole piles, but the other points (from the
present tests) are for side shear only, it is
strictly not correct to combine them into a
single empirical relationship for μ. However,
if it is accepted that μ values for end bearing
are close to unity for all soil types, then
the variation of the grey points with LL is
mainly caused by the effect of the variation
of LL on side shear. Then, for saturated soils,
the grey points can be combined with the
green points and the two of the blue points
that were for saturated conditions, to give
the best fit linear Equation 6 for μ for side
shear in saturated soil. It has R2 = 0.73.
μ = –3.895e-3.LL + 0.731 (6)
It is noted that if a μ value from Equation 6
is used to interpret a rapid test on a
whole pile (including its end-bearing
resistance where μ is close to unity) then
a conservatively low estimate of the static
load resistance will be obtained.
CONCLUSIONS
QQ Improvement in the scatter of values of μ
previously published has been obtained
by doing “static” and rapid tests on the
same pile, but repeating the static test
26
40 50 60 70
Calculated by Equation 4
after the rapid test to get a more accurate
assessment of static behaviour to which
to compare the rapid behaviour.
QQ Based on three tests, having a wide
range of soil types, the factor μ (refer to
Equation 2) for end-bearing loads on a pile
is close to 1 and independent of soil type.
QQ The factor μ for side shear in coarse
soils is confirmed to be close to 1.
QQ Based on multiple tests on 12 soils, μ for
side shear on a pile varies with: (i) soil type
(coarse or cohesive), and (ii) for a cohesive
soil, its liquid limit and its water content.
QQ Values of μ for side shear for use in
practice may be determined from the liq-
uid limit and water content of a soil using
the empirical Equation 5 in this paper, for
the range of liquid limit up to 60.
QQ In saturated soils, Equation 6 is
applicable to give μ for side shear, for
liquid limit up to 90.
QQ Conservative values of maximum “static”
capacity for a whole pile may be obtained
by applying values of μ from Equations 5
and 6 to the unloading-point method
of interpretation of a rapid test on a
whole pile.
ACKNOWLEDGEMENTS
The author wishes to thank Civilab (Pty)
Ltd for their generosity to the general
Volume 63 Number 2 June 2021
public, via the geotechnical engineering
profession, for doing all the material char-
acterisation tests reported in this paper
without charge.
The assistance of N Mangisana and
L Madlala in obtaining the end-bearing test
results during their BSc final-year investi-
gational project is gratefully acknowledged.
REFERENCES
ASTM 2008. ASTM D7383-08. Standard Test Methods
for Axial Compressive Force Pulse (Rapid) Testing of
Deep Foundations. [Superseded by ASTM D7383 –
19.] West Conshohocken, PA: ASTM International.
Brown, M J 2004. The rapid load testing of piles in fine
grained soils. Sheffield, UK: PhD thesis. University
of Sheffield.
Brown, M J & Powell, J J M 2013. Comparison of rapid
load test analysis techniques in clay soils. Journal of
Geotechnical and Geoenvironmental Engineering,
ASCE, 139(1): 152–161.
Byrne, G, Chang, N & Raju, V 2019. A Guide
to Practical Engineering in Africa, 5th ed.
Johannesburg: Franki Inc.
Fleming, W G K, Weltman, A J, Randolph, M F &
Elson, W K 1994 Piling Engineering. Glasgow:
Blackie & Son.
Hölscher, P, Brassinga, H, Brown, M J et al 2012. Rapid
Load Testing on Piles: Interpretation Guidelines.
London: CRC Press.
ISO (International Standardisation Organisation) 2016.
ISO 22477-10:2016. Geotechnical Investigation and
Testing: Testing of Geotechnical Structures. Part 10:
Testing of Piles: Rapid Load Testing. Geneva: ISO.
JGS (Japanese Geotechnical Society) 2002. JGS 1815-
2002. Method for Rapid Load Test of Single Piles.
Tokyo, Japan: JGS.
Middendorp, P, Bermingham, P & Kuiper, B 1992.
Statnamic load testing of foundation piles. In:
Barends, F B J (Ed.). Application of Stress Wave
Theory to Piles. Rotterdam, The Netherlands:
A A Balkema, 581–588.
Middendorp, P & Bielefeld, M W 1995. Statnamic load
testing and the influence of stress wave phenomena.
Proceedings, 1st International Statnamic Seminar,
25–27 September 1995, Vancouver, Canada,
pp 1–15.
SANS (South African National Standard) 1983. SANS
1200F 1983 Standardised Specification for Civil
Engineering Construction. Section F: Piling. Pretoria:
SABS Standards Division.
Weaver, T J & Rollins, K M 2010. Reduction factor
for the unloading point method at clay soil sites.
Journal of Geotechnical and Geoenvironmental
Engineering, ASCE, 136(4): 643–646.
Journal of the South African Institution of Civil Engineering
 TECHNICAL PAPER
A critical review of the
Water Sorptivity Index (WSI) Journal of the South African
Institution of Civil Engineering

parameter for potential

ISSN 1021-2019
Vol 63 No 2, June 2021, Pages 27–34, Paper 1123

durability assessment: AMY MOORE completed her BSc (Eng) and

MSc (Eng) degrees in the Department of

Can WSI be considered in

Civil Engineering at the University of Cape
Town. Her Master’s research focused on the
corrosion of reinforced concrete in marine

isolation of porosity?
environments. Since having completed her
studies, Amy has been pursuing a career in
education and publishing.

Contact details:
CoMSIRU
A J Moore, A T Bakera, M G Alexander Department of Civil Engineering
University of Cape Town
Private Bag X3, Rondebosch 7701, Cape Town, South Africa
T: +27 21 650 4012
The Durability Index (DI) approach used in South Africa is linked to transport-related properties
E: amymoore.ct@gmail.com
of the cover layer of concrete. This performance-based approach utilises three different
durability-related parameters, namely Water Sorptivity Index (WSI), Oxygen Permeability Index ALICE TITUS BAKERA is a PhD student in the
(OPI) and Chloride Conductivity Index (CCI). In the water sorptivity test, both the sorptivity Department of Civil Engineering at the
and the water-penetrable porosity of the concrete are measured. Until now, porosity has not University of Cape Town (UCT). Her current
research focuses on the biogenic acid attack
been considered directly as a separate durability-related parameter. This paper analyses these
of concrete subjected to the sewer
implications and suggests that porosity should be adopted as a further index. The paper environment. She is a member of the
concludes that sorptivity cannot be interpreted without also reflecting the porosity value since Concrete Materials and Structural Integrity
these two parameters are inter-related. Research Unit (CoMSIRU) at UCT and a
member of the Concrete Society of Southern Africa, Western Cape Branch.

Contact details:
CoMSIRU
INTRODUCTION AND It seems self-evident that measurement
Department of Civil Engineering
BACKGROUND of suitable parameters is essential to the University of Cape Town
After water, concrete is the most used achievement of durability, but the difficulty Private Bag X3, Rondebosch 7701, Cape Town, South Africa
material on earth, with over 25 billion has been the lack of appropriate durability- T: +27 21 650 4012
E: bkrali003@myuct.ac.za
tons being produced worldwide annually related tests and parameters that can be
(World Business Council for Sustainable applied, especially on as-built structures.
PROF MARK ALEXANDER is an Emeritus
Development 2009). However, the premature DIs help to fulfil this requirement. Two Professor and a Senior Research Scholar at
deterioration of concrete structures is of other critical aspects in the selection of the University of Cape Town (UCT). He has a
increasing concern around the world. In suitable durability index parameters are the PhD from the University of the
Witwatersrand and is a Fellow of the
developing countries such as South Africa, purpose for which the durability param- University of Cape Town and of RILEM
the excessive cost of repairing and rehabili- eters are measured, and the ease and speed (International Union of Laboratories and
tating concrete infrastructure is detrimental of measurement so that they can be used Experts in Construction Materials, Systems
and Structures). He co-authored Aggregates in Concrete (2005),
to the needs of other sectors of the economy. as rapid indicators of quality in quality
Alkali-Aggregate Reaction and Structural Damage to Concrete (2011) and
Resources that could be directed towards control schemes. It is this combination of Durability of concrete – design and construction (2017) (CRC Press), and
urgently needed new construction are purpose and ease and speed of measure- was Editor of Marine concrete structures. Design, durability and
diverted to the cost of repairing existing ment that dictates whether a parameter performance (2016) (Woodhead Publishers). He is involved in the Concrete
Materials and Structural Integrity Research Unit (CoMSIRU) at UCT, which
infrastructure. These resources could also can be used as a durability indicator.
focuses on infrastructure performance and renewal research.
be better spent on education and developing
Contact details:
additional infrastructure. Therefore, it is Importance of durability
CoMSIRU
essential that ways be found to ensure that design and assessment Department of Civil Engineering
concrete structures are constructed in the Durability design is integral to structural University of Cape Town
most durable way possible. To respond to design. However, durability design needs to Private Bag X3, Rondebosch 7701, Cape Town, South Africa
T: +27 21 650 4012
this challenge, researchers are developing be emphasised during the design process E: mark.alexander@uct.ac.za
new approaches and tools to address durabil- to ensure that concrete structures are
ity problems. A powerful new set of tools designed for longevity. Durability is defined
involves “durability indicators”, or “durability as the ability of a material or structure to
indexes (DIs)” which are linked with the pre- withstand the service conditions for which
ferred performance-based approach to dura- it was designed, without undue deteriora-
bility design and specification. These DIs are tion (Alexander & Beushausen 2019). As Keywords: concrete durability, durability index, water sorptivity,
discussed later (Alexander et al 2008). indicated, many concrete structures are porosity, absorption

Moore AJ, Bakera AT, Alexander MG. A critical review of the Water Sorptivity Index (WSI) parameter for potential durability assessment:
Can WSI be considered in isolation of porosity? J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #1123, 8 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a4 27
showing inadequate durability, implying
that the objectives of “durability design”
have not been particularly successful. To
help remedy the premature deterioration
problem, “potential durability” of the con-
crete needs to be assessed and specified.
Potential durability is taken here to mean
the ability of the concrete delivered to, or
mixed on, the job site to be durable in the
structure for which it is intended, under
the given exposure conditions.
Many of the deterioration processes
that render reinforced concrete structures
vulnerable are concerned with the quality
and depth of the concrete cover to the rein-
forcing steel. Potential durability relates to
the quality of the cover layer (which is a
complex product of the concrete constitu-
ents and proportions) and site construction
processes, including protection and curing
immediately post-construction. These
all have a direct influence on the rate
of ingress of deleterious substances into
the concrete. Cover quality, with a direct
impact on potential durability, can be
accomplished by ensuring that appropri-
ate materials are selected for the given
environment at the design phase, and that
construction processes such as compaction
and curing are correctly carried out.
Prescriptive versus performance
approach to durability
design and specification
Traditionally, durability design and
specification were achieved by applying
prescriptive approaches, in which construc-
tion materials and other constraints (e.g.
water/binder ratio, concrete cover, and
grade of concrete) were adjusted to suit the
environmental exposure class. However,
this approach is now much criticised as it
does not consider (a) the ability to measure
actual potential durability “resistance”, and
(b) the effects of the construction processes
which affect durability, mainly curing and
compaction. Further, modern concrete-
making materials, specifically binders and
admixtures, can no longer be regarded as
“simplistic” and need a more sophisticated
approach to characterise and use them
properly. Therefore, performance-based
approaches, which must require the quality
of the as-built structure to be assessed to
be truly performance-based, are crucial in
achieving durable structures. The cover
zone of the as-built structure needs to
be tested to reflect the penetrability and
thus potential durability of the concrete.
While the composition of the constituent
28
materials and mix proportions are impor-
tant, ultimately, it is equally important to
quantify in situ performance (Alexander &
Beushausen 2019).
Quantification of durability
From the above, it follows that the best
durability strategies for reinforced
concrete structures will be those that
seek to improve the quality of the cover
concrete. This quality must be appropri-
ately measured, during construction, as
well as the actual cover depth achieved,
thus quantifying potential durability.
Potential durability, as defined above, is
readily quantified by means of durabil-
ity indexes. This concept can be taken
further by linking it to service-life models
(Mackechnie & Alexander 2002) which can
then be used to predict deterioration of a
structure. Durability should be quantified
by considering indexes that represent the
transport properties of concrete, while also
considering the aggressiveness of the expo-
sure environment. This latter point is very
important, since it was observed by Moore
(2014) that reinforced concrete structures
in the marine environment, of the same
quality concrete, had higher rates of cor-
rosion in the splash zone, when compared
with the tidal zone.
DURABILITY INDEX APPROACH
The durability index approach has been
developed in South Africa (SA) over recent
decades (Alexander et al 2018, 2008, 1999;
Mackechnie & Alexander 2002). The SA
durability indexes are measurable durabil-
ity parameters and serve as “indexes” of
the potential durability of the material or
structure. They characterise the key mate-
rial properties that govern the durability
issues of concern. Although being meas-
ured in the short-term (typically 28 days),
research has shown that they can be used
as indicators of the likely durability per-
formance of the structure in the long-term
(Alexander & Beushausen 2019).
As discussed, the indexes reflect the
potential durability of the concrete. This
can be affected if concrete is cast and
cured under laboratory conditions, in
which case the indexes generally (but not
always) represent concrete quality that is
superior to both as-delivered quality, and
as-built quality, with the latter reflecting
the influence of construction practices.
Thus, it is possible to assess whether the
concrete can be expected to be durable
Volume 63 Number 2 June 2021
under “ideal” conditions, as well as actual
concrete under site conditions. Since site
conditions are never “ideal”, allowance
must be made in design for exposure con-
ditions and construction effects.
Brief description of Durability
Index tests
The South African DI tests consist of
three tests, namely, Oxygen Permeability
Index (OPI), Chloride Conductivity Index
(CCI) and Water Sorptivity Index (WSI)
tests. These tests have been discussed
in literature and are briefly described
here (Alexander et al 2018, 2008, 1999;
Mackechnie & Alexander 2002). All the
tests are conducted on “standard” 70 mm
diameter, 30 mm thick specimens. The
specimens are pre-conditioned by oven-
drying and, for the CCI test, also by pre-
saturation with chloride solution.
The OPI test measures the pressure
decay of oxygen passed through a standard
specimen in a falling head permeameter,
the index being defined as the negative
log of the coefficient of permeability.
Common OPI values range from 8.5 to 10.5
(on a log scale), a higher value indicating
a higher impermeability and thus a con-
crete of potentially higher quality. Using
this approach, carbonation depths may
be predicted for different environments
(Salvoldi et al 2015).
The CCI test measures the instantane-
ous electrical current due to the flux of
chloride ions through a standard specimen,
pre-saturated in 5 M sodium chloride
(NaCl) solution, in a two-cell conduction
rig in which both sides of the specimen are
exposed to the NaCl solution. The chloride
conductivity index (CCI) is related to the
chloride diffusion properties of the concrete.
Finally, the WSI test measures the rate
of movement of a wetting front through
concrete under capillary suction. A
schematic presentation of the test setup
is given in Figure 1 – the lower the WSI
value, the better the potential durability
of the concrete. This index is particularly
sensitive to the near-surface properties of
concrete and is, thus, a good indication of
the type and effectiveness of curing. As the
final step in the test, the water-penetrable
porosity is measured under vacuum. The
WSI parameter is in effect the rate of mass
change from water uptake by absorption,
divided (or “normalised”) by the water-
penetrable porosity of the specimen. This
has implications for the resulting sorptivity
values, which will be explored later.
Journal of the South African Institution of Civil Engineering

Concrete core specimen
70 mm diameter
30 mm thick
Epoxy coating
Capillary rise
Test surface
Figure 1 S chematic representation of the WSI test
Table 1 Suggested range for durability classifications using index values (Alexander et al 1999)
Durability class OPI (log scale)
Excellent >10
Good 9.5–10
Poor 9.0–9.5
Very poor <9
These tests are useful since they are
rapid and can be conducted typically at
28 days after construction, allowing an
early age evaluation of potential durability
(Alexander et al 2008). Since the original
implementation of the tests, they have been
improved and updated (Mukadam et al
2016; Otieno & Alexander 2015).
Matrix of DI values
The durability indexes have been placed
in a matrix of values which can be used to
evaluate the potential durability of concrete
structures at an early age (see Table 1). The
table is qualitative and therefore has limita-
tions in its applications since the level of
acceptable “performance” must be consid-
ered in relation to the exposure conditions
and service environment of the structure.
Table 1 can be viewed as overly simplistic
because it does not directly consider the
latter two factors; for this reason more
sophisticated specifications have been
developed by the South African National
Roads Agency Limited (SANRAL) that take
these into account explicitly (but beyond
the scope of this paper) (SANRAL 2009).
Durability studies must also address the
relation of indexes to fundamental concrete
properties and microstructure. The work
presented in this paper attempts such by
considering the specific durability proper-
ties of sorptivity and porosity, and how these
are related in terms of potential durability.
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
Wet paper towels
Sorptivity (mm/√h ) Conductivity (mS/cm)
<6 <0.75
6–10 0.75–1.5
10–15 1.5–2.5
>15 >2.5
SORPTIVITY VERSUS POROSITY
In the WSI test, both the sorptivity and
the water-penetrable porosity of the
concrete are measured. The sorptivity
value is effectively the rate of mass gain
(g/√h) “normalised” by the porosity of the
concrete. The index is expressed as a rate
of penetration depth of a wetting front
into the concrete, typically in mm/√h,
and not purely as the rate of mass uptake.
For deterioration mechanisms such as
reinforcement corrosion, the rate at which
fluids approach embedded reinforcing bars
is critical in evaluating the expected service
life. Generally, concrete sorptivity is gov-
erned by a square root of time law, with the
relevant expressions given in Equations 1
and 2:
Fd
S= (1)
Msv – Mso
where S (mm/√h) is sorptivity; F (g/√h is
defined in Equation 2; Msv (g) is the satu-
rated mass of the specimen; Mso (g) is the
dry mass of the specimen; and d (mm) is
the thickness of the specimen.
Mwt = F√t(2)
where Mwt (grams) is the mass of the
sample, and F is the slope of the best fit
line from plotting Mwt against the square
root of time t; t being the time (typically in
hours) after a specimen is first exposed to
water on its wetting face.
The denominator term (Msv – Mso) in
Equation 1 is a proxy for porosity (Nilsson
2003) and thus Equation 1 normalises
the measured rate of mass gain “F” by a
porosity-related term, to give the sorptivity
S. Porosity n (in percentage) is defined in
Equation 3:
Msv – Mso
n= × 100(3)
Adρw
where Msv, Mso and d are defined above,
A is the cross-sectional area of the speci-
men, and ρw is the density of water. [Note
that the term (Msv – Mso) appears in
the numerator in Equation 3, and in the
denominator in Equation 1; it is in this
sense that the sorptivity can be thought of
as the rate of mass gain “normalised” by
the porosity.]
In fact, water-penetrable porosity is
the fundamental material parameter that
is measured in the WSI test, not sorptiv-
ity. Porosity is a complex parameter that
encompasses pore volume, pore intercon-
nectivity, pore tortuosity, and pore path
length. In the test it is assumed that all
pore volumes that are penetrable by water
are filled, giving a mass change that can
be converted into a percentage volume
of the specimen. This, together with the
measured rate of mass change due to water
absorption, gives the sorptivity. This is
illustrated schematically in Figure 2, where
the measured values are distinguished
from the inferred value of sorptivity.
Combining Equations 1 and 3 results in
Equation 4, which indicates that sorptivity
and porosity are related, showing that two
specimens with the same measured rate of
mass gain can have different sorptivity val-
ues if their porosities differ, and vice versa.
F
Sn = × 100(4)
Aρw
For two different concretes (subscript 1 and
2) with the same rate of mass change, the
F
term × 100 is constant, indicating that
Aρw
the products of their sorptivity and poros-
ity values are equal (see Equation 5).
S1n1 = S2n2(5)
It is the inter-relation between these two
durability index parameters that is of
29

Measured
Porosity
(percentage water–penetrable
volume) (%)
Uni-directional mass rate of
change due to absorption (g/√h)
Figure 2 M
 easured versus inferred parameters in the WSI test
interest here, with the question of whether
both parameters should be considered in
durability assessments, and if so, what rela-
tive importance should be given to each.
Porosity as a separate
durability indicator
Other sorption-type tests, e.g. ASTM
C1585-13 (ASTM 2013), are based on
absorption only (amount of water absorbed
over the surface area), which might be con-
sidered of limited value. For example, for
concrete with high porosity, high absorp-
tion may relate to low penetration depth
and potentially acceptable penetrability
properties (i.e. low depth of ingress during
a wetting cycle). However, such a measure
does not account for the possibility that
a lot of fluid might be absorbed into the
surface of the concrete, and, if contami-
nated (e.g. by chlorides), may result in high
contaminant concentrations at the surface,
resulting in low durability (e.g. from dif-
fusion of chlorides). Thus, a measure of
porosity, representing the volume of acces-
sible pores in the cover zone of concrete, is
needed as well.
Until now, porosity has not been con-
sidered directly as a separate durability
parameter amid the suite of DI param-
eters. However, sorptivity and porosity
are inversely related (Equation 4), and
in laboratory measurements it has been
noticed that there can be high or low
values of sorptivity associated with high
or low values of porosity, and vice versa.
This brings into question whether poros-
ity should be considered in conjunction
with sorptivity as a durability parameter,
and therefore also the issue of what the
acceptable values of these two parameters
30
Inferred
Sorptivity
(rate of advance of wetting
front) (mm/√h)
in combination should be. This issue needs
to be addressed, so that durability tests are
fully understood, and to improve predic-
tions of concrete performance.
This paper analyses the implications of
the above and suggests that porosity should
be adopted as an index; importantly, the
paper recommends that sorptivity cannot
be interpreted in the absence of porosity
since these two parameters are related and
their interpretation regarding deteriora-
tion mechanisms needs to consider both
parameters.
OBJECTIVE OF THIS PAPER
The objective of this paper is, therefore, to
explore differing values of porosity for the
same sorptivity, based on extensive previous
experimental results, and what implications
this might have for the potential durability
of concrete. However, at this stage, potential
durability has not been quantified by way of
any service-life model, since porosity is also
important in service-life predictions, imply-
ing that a matrix of suitable values would
have to be determined and provided.
METHODOLOGY
For examining the sorptivity–porosity rela-
tionship, data was utilised from our availa-
ble laboratory database, in which sorptivity
and porosity values for various specimens
had been evaluated over an extended
period. A total of 601 data points, obtained
from different specimens from 2009 to
2015, tested as per Alexander et al (2018),
were collated and analysed. Each data point
consisted of sorptivity and porosity for a
given specimen. A limiting factor of this
Volume 63 Number 2 June 2021
data was that, sometimes, details of the
concrete mixes were missing. Information
such as the compressive strength, w/c ratio
and binder type would have been useful in
making other inferences.
As shown in Equations 1 and 2, in the
absence of data on the rate of mass change,
it is possible to obtain this parameter
from the sorptivity, knowing the porosity
(in effect the term (Msv – Mso)). As some
information sources that were analysed
did not have the rate of change of mass
recorded, the values were determined indi-
rectly as indicated.
It was important to establish the rate
of change of mass F of the different data
points so that a comparison could be made
between sorptivity and porosity for the
same rate of mass change. The rate of mass
change is also useful in establishing con-
crete durability. It stands to reason that a
concrete with a higher rate of mass change
should have a lower durability potential,
because a higher rate means that more
water (and therefore potentially harmful
substances) may enter the concrete. This
increases the probability of corrosion or
other deterioration occurring.
RESULTS AND ANALYSIS
Inter-relationship of
sorptivity and porosity
The sorptivity and porosity values from 601
data points recorded in previous research
were plotted on a scatter plot (Figure 3).
Once the outliers were removed from
the data, the correlation coefficient (R2)
between sorptivity and porosity was
evaluated as 0.19. As the data set is large
and consists of specimens of different mix
designs and constituent materials, this is
to be expected. However, a general trend
is observed: higher sorptivity values tend
to have higher porosity values. This might
be somewhat surprising, considering the
earlier argument on the expected inverse
relationship between these two parameters.
However, it is also possible to have high
sorptivity and corresponding high porosity
values if the pores are highly intercon-
nected and small in size. This is largely a
consequence of taking diverse data and
combining them into a single data set. As
will be shown, in general, sorptivity and
porosity tend to be inversely related.
Box and whisker diagrams of all the
sorptivity and porosity values in the data
set were produced, in Figures 4 and 5
Journal of the South African Institution of Civil Engineering
 data represented concrete grades of 50,
18 40, 30 and 25 MPa. Following this, data
R2 = 0.19 points with similar rates of change of mass
16
(parameter F in Equation 2) were grouped
14 together, and the corresponding sorptivity
and porosity values were plotted against
Porosity (%)

12 each other. The categories of the groups

of rates of mass change were taken as 6–7,
10
7–8, 8–9, and 9–10 g/√h.
8 Figure 6 shows scatter plots of the
sorptivity and porosity values plotted for
6 different mass rates of change. The same
4
pattern can be observed across all four
2 4 6 8 10 12 14 16 scatter plots, with the sorptivity and poros-
Sorptivity (mm/√h) ity values being closely inversely related
for the same category of rate of change of
Figure 3 D
 ata sets (601) represented in a scatter plot: porosity versus sorptivity mass. The plots show that, for any given
absorption rate, the relationship between
sorptivity and porosity is essentially the
same regardless of concrete grade. This is
an important observation since it indicates
that the same sorptivity value is possible
for different grades of concrete, governed
largely by the different rates of change of
mass. The data also illustrates that for a
given concrete or closely related set of con-
cretes, low sorptivity values can result from
high porosity values and vice versa.
Further observation of the four plots
in Figure 6 shows that, as the rate of mass
0 2 4 6 8 10 12 14 16 18 change increases, the scatter bands move up
Sorptivity (mm/√h) and to the right in relation to the axes, imply-
ing that both sorptivity and porosity increase
Figure 4 B
 ox and whisker diagram showing distribution of sorptivity values with increase in rate of mass change, i.e.
complementing Equation 4. This is expected
for sorptivity, since it relates directly to rate
of mass change; however, the fact that poros-
ity also increases implies that the rate of mass
change increases faster than the increase in
pore volume. This is most likely linked to
greater pore interconnectivity at greater rates
of mass change. Both these aspects – greater
sorptivity and greater porosity at greater rate
of mass change – are undesirable for produc-
ing a durable concrete. This also intimates
that both parameters need to be considered
when assessing the potential durability of a
0 2 4 6 8 10 12 14 16 18 concrete mix.
Porosity (%) The scatter band plots in Figure 6
represent concrete strength grades over
Figure 5 B
 ox and whisker diagram showing distribution of porosity values a wide range of 25–50 MPa. While there
is some clustering of points for a given
respectively, showing a symmetrical distri- Analysis of data strength grade, the picture emerges that
bution of the data. The average sorptivity the two durability-related parameters are
value of the data set is 8.5 mm/√h, while Data categorised according to rate of mass not meaningfully linked to strength. This
the range of sorptivity is 3–14 mm/√h. The change, incorporating strength grade is also an important conclusion since many
porosity values are also found to be sym- Data points that had information pertain- engineers still relate durability to strength.
metrical, with the average porosity being ing to the stated grade of the concrete These results show that it is possible to
11.9%, and the range of porosity values (not measured grade), of which there were obtain a given sorptivity or porosity over a
being 5–17%. 151, were extracted and categorised. The wide range of concrete strengths.

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 31
 6–7 (g/√h) 7–8 (g/√h)
10.5 15

10.0 14

9.5 13
Sorptivity (mm/√h)

Sorptivity (mm/√h)
9.0 12

8.5 11

8.0 10

7.5 9

7.0 8

6.5 7

6.0 6
9 10 11 12 13 14 15 16 8 9 10 11 12 13 14 15 16 17
Porosity (%) Porosity (%)
50 MPa 40 MPa 30 MPa 25 MPa 50 MPa 40 MPa 30 MPa 25 MPa

8–9 (g/√h) 9–10 (g/√h)

13 15

14
12
13
11
Sorptivity (mm/√h)

Sorptivity (mm/√h)

12
10 11

9 10

9
8
8
7
7

6 6
8 9 10 11 12 13 14 15 16 17 8 9 10 11 12 13 14 15 16 17
Porosity (%) Porosity (%)
50 MPa 40 MPa 30 MPa 25 MPa 40 MPa 30 MPa 25 MPa

Figure 6 S orptivity and porosity relationships for different categories of rate of change of mass, over a range of compressive strength grades

Relationship between sorptivity and rate Table 2 R

of change of mass for a constant porosity
In general, an increase in the rate of water
absorption will result in a decrease in
potential durability. Regarding Figure 6, it is
possible to obtain the same sorptivity over
a range of values of rate of change of mass,
moderated again by the porosity. For exam-
ple, a sorptivity of 10 mm/√h is associated
with a range of porosities, depending on the
rate of change of mass, given in Table 2.
It can be argued that concrete with a
sorptivity of 10 mm/√h will have a greater
potential durability at a porosity of 10%
than at a porosity of 15%, particularly since,
as indicated above, the higher porosities
may represent more interconnected pores.
Thus, sorptivity cannot be considered in
isolation of porosity since both can influ-
ence potential durability.
 ate of change of mass and associated porosity, for a sorptivity of 10 mm/√h
(data from Figure 6)
Rate of change of mass category (g/√h)
Approximate porosity (%)
It is also important to consider the
types of degradation that might occur
because of different environmental exposure
conditions, as well as the function of the
structure, as these play a role in whether
sorptivity or porosity is more important. For
freeze-thaw degradation, the porosity and
critical pore size are probably more crucial
than the sorptivity, due to the mechanism
of freezing which involves the expansion
of water in the pores. The larger the pore
size, the larger the total amount of liquid in
Sorptivity of 10 mm/√h
6–7 7–8 8–9 9–10
10.0 11.5 13.5 15.0
the pore available to expand. The amount
of space and the pore spacing factor are
therefore critical in determining the amount
of damage. For other types of degradation,
such as corrosion of reinforcement, sorptiv-
ity is more crucial. If the sorptivity is higher,
concrete exposed to a fluid will experience
deeper wetting in a shorter period, with the
fluid reaching the reinforcing bar and influ-
encing corrosion. This might be critical in,
say, zones of wetting and drying such as the
inter-tidal zone in marine structures.

32 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 Very poor (>15)
14 to 16 8.0 8.8 9.6 10.5 11.1 12.0 12.9 13.8 14.5 14.9
11 13
12 to 14 7.0 7.5 8.0 9.5 10.1 11.0 12.0 12.9 13.6 14.5
Poor (10–15)
Sorptivity (mm/√h)

9
10 to 12 6.0 6.8 7.4 8.0 8.6 9.2 9.8 10.4 11.4 12.2
7

8 to 10 5.2 5.8 6.1 6.6 7.0 7.6 8.4 8.7 9.5 10.0
5 Good (6–10)
6 to 8 3.9 4.4 4.9 5.5 6.0 6.4 6.8 7.4 7.9 8.3

3
4 to 6 2.7 3.1 3.5 3.8 4.3 4.5 5.0 5.5 6.0 6.5 Excellent (<6)

7 to 8 8 to 9 9 to 10 10 to 11 11 to 12 12 to 13 13 to 14 14 to 15 15 to 16 16 to 17
Porosity (%)

Values in white boxes on the contours represent rate of mass change in g/√h.

Figure 7 M
 atrix of porosity, sorptivity and associated rate of change of mass (in g/√h)

Semi-quantitative matrix of
sorptivity and porosity values
The main objective of this work was to
determine whether sorptivity should be
considered in conjunction with porosity
when assessing the potential durability
of concrete. The question that arises is:
what porosity values in combination with
sorptivity values might be considered
“acceptable” or not? To assist with mak-
ing this assessment, a semi-quantitative
matrix of values of sorptivity and porosity
in combination was produced. The ranges
of sorptivity and porosity were selected
as 4–16 mm/√h and 7–15% respectively.
A gradation of 2 units was used for the
sorptivity axis, and a gradation of 1 unit for
the porosity axis. The matrix is shown in
Figure 7.
Values of the average rate of mass
increase for possible combinations of sorp-
tivity and porosity were inserted on Figure 7,
associated with “contours” (see white boxes).
A colour scale was used to show high (red)
and low (green) rates of change of mass. It
is reasonable to assume that the higher the
water absorption rate, the more vulnerable
the concrete will be in terms of durability.
From Figure 7, for a given sorptivity
value, there exists a wide range of rate of
change of mass values. This shows again
how concretes with the same sorptivity
value can have varying potential durability
depending on the porosity. Concretes with
higher porosity might be more penetrable
to greater depth over the long term than
concretes with lower porosity. The con-
tours attempt to show that concretes with
any combination of sorptivity and porosity
values that fall on the same contour line,
or within a set of contours, may have judgement was used to impose limits on
similar potential durability (given the same the values, such that both sorptivity and
environmental conditions). For example, porosity values of greater than 15% were
in this scheme a concrete with a sorptivity considered undesirable in a structural con-
of 6 mm/√h and a porosity of about 12% crete. The desired durability classifications
may be considered to have similar potential should also take into consideration the envi-
durability as a concrete with a sorptivity of ronmental exposure class of the structure.
8 mm/√h and a porosity of about 8%. A “poor” concrete might be quite acceptable
Figure 7 also gives a qualitative and in an environment where the structure is
semi-quantitative classification of durabil- not exposed to aggressive substances, such
ity in the sorptivity categories on the right- as internal concrete in a normal building.
hand side of the diagram, which derive Thus, engineers and specifiers should exer-
from Table 1. Conventionally, these catego- cise caution in using Table 3, so as not to
ries are extended leftwards by following over-specify a given concrete.
the relevant horizontal category lines from
their starting positions on the right-hand
side axis. This is the more conservative CONCLUSIONS
approach. However, an alternative scheme
that bears further investigation is to extend Effect of strength grade on
the sorptivity categories leftwards by the relationship between
following the relevant contour lines. For sorptivity and porosity
example, any concrete falling under the For any given absorption rate (i.e. rate of
mass rate of gain contour of, say, 7 g/√h mass change), the relationship between
might be considered “excellent”, together sorptivity and porosity is essentially the
with other concretes within the contour same regardless of the nominal grade of
limits with different porosities and sorptiv- the concrete (50, 40, 30 and 25 MPa). This
ity. It must be stressed that these potential
durability classifications are tentative and Table 3 S uggested durability classification
can only be taken in a general sense. They table for sorptivity and porosity values
are presented here to stimulate debate on Sorptivity Porosity Durability
how these parameters might better be used (mm/√h) (%) classification
in durability design.
<6 <10 Excellent
To make the matrix more “user-friendly”,
the results were simplified into Table 3 that <10 Excellent to Good
6–10
gives the potential durability for a combina- >10, <12 Good to Poor
tion of different sorptivity and porosity
values. These critical values for porosity <12 Good to Poor
10–15
were determined by using the contour lines >12, <15 Poor to Very poor
of Figure 7 as a guide, but also considering
>15 – Very poor
the data in Figure 3. Importantly, however,

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 33
demonstrates that the same sorptivity value
is possible for different grades of concrete,
and highlights the importance of consider-
ing a parameter such as porosity to fully
describe the durability potential.
Also, both sorptivity and porosity
increase with increase in rate of mass
change. While this direct relationship
between sorptivity and rate of change of
mass is to be expected, it is not intuitively
expected for porosity, but implies that the
rate of change of mass increases faster
than the pore volume increases. This
leads to the conclusion that higher rate
of mass change is linked to greater pore
interconnectivity (and not necessarily pore
volume). An increase in both or either of
these parameters, sorptivity and porosity,
will result in a decrease in the potential
durability of the concrete. For this reason,
both sorptivity and porosity values should
be considered when assessing the potential
durability of concrete.
The results also show that a given
concrete strength could have varying com-
binations of porosity and sorptivity. This
confirms the notion that strength is not
indicative of durability, in contrast to what
is commonly thought by engineers.
Porosity considered in
conjunction with sorptivity
As mentioned above, concretes with
the same sorptivity may have varying
porosity values (an increase in porosity
is associated with an increase in the rate
of mass change). It can thus be argued
that concrete with a given sorptivity rate
and “lower” porosity, will have a greater
potential durability than a concrete with
a “higher” porosity, since higher porosities
represent greater connectivity of pores.
Thus, sorptivity cannot be considered in
isolation to porosity, as both parameters
influence the potential durability.
To attempt to quantify this, a table of
suggested “acceptable” combinations of
34
sorptivity and porosity values was devel-
oped (Table 3). Judgement was used to
impose limits on very high porosity and/
or sorptivity values. Sorptivity and porosity
values greater than 15% were considered
undesirable. While this table is useful in a
general sense, the environmental exposure
conditions of the structure need to be
considered when assessing the potential
durability. For example, a “poor” sorptiv-
ity classification may be acceptable for a
concrete that forms part of the interior
of a normal building, where deterioration
is extremely unlikely. Further research is
needed to refine the table of “acceptable”
porosity and sorptivity values and explore
the implications for concrete potential
durability. Also, further work is needed
to assess the effect of measured porosity
values and pore size distributions on the
durability parameters considered.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the
financial support of the Concrete Materials
and Structural Integrity Research Unit
(CoMSIRU) at the University of Cape Town,
and the industrial sponsors of CoMSIRU,
i.e. PPC Cement (Pty) Ltd, AfriSam, Lafarge,
SIKA, Haw & Inglis, and AVENG. Other
sponsors of CoMSIRU are the Water
Research Commission of South Africa
(WRC), the South African Department of
Water and Sanitation, and Transnet.
REFERENCES
Alexander, M, Ballim, Y & Mackechnie, J 2018. Durability
Index Testing Procedure Manual. Version. 4.5.1.
CoMSIRU. University of Cape Town & University of
the Witwatersrand. http://www.comsiru.uct.ac.za.
Alexander, M G, Ballim, A Y, Stanish, K & Ballim, Y
2008. A framework for the use of durability indexes
in performance-based design and specifications
for reinforced concrete structures. Materials and
Structures, 41(5): 921–936.
Volume 63 Number 2 June 2021
Alexander, M G & Beushausen, H 2019. Durability,
service life prediction, and modelling for reinforced
concrete structures: Review and critique. Cement
and Concrete Research, 122: 17–29.
Alexander, M G, Mackechnie, J R & Ballim, Y 1999.
Guide to the use of durability indexes for achieving
durability in concrete structures. Resesearch
Monograph No. 2. University of Cape Town &
University of the Witwatersrand.
ASTM 2013. ASTM C1585-13 2013. Standard Test
Method for Measurement of Rate of Absorption
of Water by Hydraulic Cement Concretes. West
Conshohocken, PA: ASTM International.
Mackechnie, J R & Alexander, M G 2002. Durability
predictions using early-age durability index testing.
Proceedings, 9th Durability and Building Materials
Conference, 17–21 April 2002, Brisbane, Australia,
p 11.
Moore, A 2014. Effect of oxygen availability on the
corrosion rate of reinforced concrete in marine
exposure zones: Inference from site and lab studies.
MSc dissertation. University of Cape Town.
Mukadam, Z, Alexander, M G & Beushausen, H D
2016. The effect of drying preconditioning on the
South African durability index tests. Cement and
Concrete Composites, 69: 1–8.
Nilsson, L O 2003. Durability concept: pore structure
and transport processes. In Newman, J & Choo, B S
(Eds). Advanced Concrete Technology. Cambridge,
UK: Woodhead Publishing.
Otieno, M & Alexander, M 2015. Chloride conductivity
testing of concrete. Past and recent developments.
Journal of the South African Institution of Civil
Engineering, 57: 55–64.
Salvoldi, B G, Beushausen, H & Alexander, M G 2015.
Oxygen permeability of concrete and its relation to
carbonation. Construction and Building Materials,
85: 30–37.
SANRAL (South African National Roads Agency
Limited) 2009. Spreadsheet: Table 6000/1 : Concrete
Durability Specification Targets (Civil Engineering
Structures only). Pretoria: SANRAL.
World Business Council for Sustainable Development
2009. The Cement Sustainability Initiative (CSI).
Recycling of Concrete. http://www.wbcsdcement.org.
Journal of the South African Institution of Civil Engineering
South African infrastructure TECHNICAL PAPER
condition – an opinion Journal of the South African
Institution of Civil Engineering

survey for the SAICE ISSN 1021-2019

Vol 63 No 2, June 2021, Pages 35–46, Paper 1030

Infrastructure Report Card DR FREDERIK CHRISTOFFEL RUST (Pr Eng,

MSAICE) has 37 years’ experience in
research in infrastructure and research
management, including fields such as roads,
F C Rust, K Wall, M A Smit, S Amod transport, water infrastructure and
buildings. He has published more than 60
peer-reviewed conference papers,
accredited journal articles and book
The South African Institution of Civil Engineering (SAICE) Infrastructure Report Cards of 2006, chapters. He is currently the managing director of PERC (Pty) Ltd.
2011 and 2017 reported that much of the South African infrastructure is in a poor condition.
Contact details:
To augment the recent 2017 Report Card, a survey was, for the first time, conducted amongst PERC
SAICE members to obtain their opinion on the condition of infrastructure (in terms of a grading), No 8 Silver Sands, Equestria, Pretoria 0001, South Africa
the trend of the condition over time, as well as the reasons for the individual’s grading. The T: +27 82 447 6098, E: chris@perc.co.za

669 respondents indicated that, apart from a few exceptions such as national airports and the
Gautrain, much of the infrastructure is in a poor state, which is very similar to the results obtained DR KEVIN WALL (Pr Eng, HFSAICE, FSAAE), a
civil engineer and town planner, is
from the SAICE Report Card process. The results were also analysed per province and indicated Extraordinary Professor at the University of
that, in the opinion of respondents, specific provinces in particular are struggling with poor Pretoria, and a non-executive board member
infrastructure. The main reasons given for the poor infrastructure were a lack of maintenance, of the City of Ekurhuleni’s wastewater
treatment entity. A past president of the
lack of institutional capability, lack of sufficient funding and over-loading of infrastructure.
South African Institution of Civil Engineering
(SAICE), he has received both the SAICE Gold
Medal and the Lifetime Award of the National Science and Technology
INTRODUCTION are due to a number of factors, including Forum. Much of his work over the last two decades has been on the
effectiveness of government spending on infrastructure, and on ways to
Well-functioning infrastructure and an insufficient funding to manage, plan and
improve the quality, reliability and sustainability of that infrastructure.
efficient built environment are essential to maintain the infrastructure assets; a short-
Contact details:
socio-economic development and poverty age of skilled resources leading to problems
Faculty of Engineering, Built Environment and Information Technology
alleviation (Perkins 2011). This is empha- with institutional capacity; and a lack of University of Pretoria
sised in a number of official government appropriate technological solutions for the Pretoria 0002, South Africa
documents, including the Medium Term problems experienced with infrastructure T: +27 82 459 3618, E: kevin.wall@up.ac.za

Expenditure Framework (SA Treasury planning, materials, design, construction,

2017), the Diagnostic Report (National
Planning Commission 2011), the National
Development Plan (NDP) (National
Planning Commission 2012), and the
National Infrastructure Plan (Presidential
Infrastructure Coordinating Commission
2012) from which flows the Strategic
Infrastructure Projects (SIPs) coordi-
nated by the Presidential Infrastructure
Coordinating Committee (PICC). Over the
next three-year period more than R900
billion in public funding alone has been
budgeted for infrastructure such as roads,
energy generation plants, water infrastruc-
ture and public buildings.
However, the NDP and the Diagnostic
Report, as well as the South African
Institution of Civil Engineering (SAICE)
Infrastructure Report Cards completed
in 2006, 2011 and 2017 (SAICE 2006;
2011; 2017) have all reported that much
of the South African infrastructure is in
a poor condition, particularly in the areas
of health, water, sanitation, and second-
ary and tertiary roads. These problems
maintenance and operation.
In addition to the above, the Fourth
Industrial Revolution (Schwab 2017) is
currently changing the nature of many
industries through the advent of new tech-
nologies. This phenomenon is driven by
technologies such as autonomous vehicles,
3D printing (additive manufacturing),
advanced robotics, new materials, the
Internet of Things, sensor technology, etc.
All of these drivers will also influence the
infrastructure provision and operations
spheres. Much infrastructure is designed
to last for long periods (20 to 50 years
and more), and therefore planners and
designers need to take these trends into
consideration both in the intrinsic nature
of infrastructure (e.g. smart materials and
embedded sensors), as well as in provid-
ing a conduit for other technologies (e.g.
housing fibre-optic cable and sensors). So,
in addition to improving the quality of
current infrastructure, South Africa will
also need to position infrastructure for the
challenges of the future.
MICHELLE ANDRE SMIT has BSc Zoology and
BEng Civil Engineering degrees from the
University of Pretoria, and is currently busy
with an Honours degree in Pavement
Engineering from the same university. She is
employed as a graduate-in-training at the
Council for Scientific and Industrial Research
(CSIR) working towards professional
registration with the Engineering Council of South Africa.
Contact details:
Graduate in Training
Council of Scientific and Industrial Research
PO Box 395, Pretoria 0001, South Africa
T: +27 84 645 4829, E: msmit3@csir.co.za
SALIM AMOD (Pr Eng, HFSAICE, FSAAE,
FAArb, Convenor SAICE Infrastructure
Report Cards) was president of SAICE in
2006. He is the initiator of the SAICE
Infrastructure Report Card and has been
convenor and a principal author of all
editions of that report to date. In 2017 he
was accorded the rare honour of the SAICE
Gold Medal. He practises in dispute resolution.
Contact details:
Postnet Suite 200, Private Bag X18, Milnerton 7435, South Africa
T: +27 83 377 0729, E: amod.sam@gmail.com
Keywords: infrastructure condition, SAICE Infrastructure Report Card,
infrastructure grading system

Rust FC, Wall K, Smit MA, Amod S. South African infrastructure condition – an opinion survey for the SAICE Infrastructure Report Card.
J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #1030, 12 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a5 35
 This paper describes the survey
which was undertaken in parallel to the
largely desktop preparation of the SAICE
Infrastructure Report Card (IRC) of 2017
(SAICE 2017). This survey was conducted
amongst SAICE members to gather their
professional view of the condition of
infrastructure in a number of sectors. The
paper focuses particularly on the results of
the survey, and also discusses some of the
reasons given for respondents’ gradings.
LITERATURE
Management and planning of infrastruc-
ture often depend on multiple criteria
decision support systems that allow for
prioritisation of maintenance and rehabili-
tation (Scholz et al 2015; Šelih et al 2008;
Woodward et al 2019). This requires the
measurement of the performance of and
the assessment of the condition of the
infrastructure (Ahluwalia 2008). These
assessments can be categorised as follows:
QQ Visual inspections
QQ Non-destructive testing
QQ Photographic and optical evaluation
QQ Data recorded from sensors and smart
sensors.
Nagarajaiah and Erazo (2016) stated that
the majority of civil infrastructure is
evaluated through visual assessment using
non-destructive methods. However, these
traditional approaches are time-consuming
and may result in hidden damage not being
detected (Ellingwood 2005; Frangopol &
Liu 2007). Ahluwalia (2008) found that
visual inspections are preferable for the
assessment of buildings. Building assets
have multiple components with a variety
of requirements, and therefore other meth-
ods are less effective. Lenett et al (1999)
evaluated the use of “rapid” multi-reference
impact testing for condition assessment of
a commissioned steel-string bridge. They
concluded that this method is useful to
detect existing damage or change in the
condition of the bridge, but that visual
inspection was still required to determine
the nature of the change and which
component was affected. In spite of their
time-consuming nature and cost, visual
inspections remain an integral part of con-
dition assessment of infrastructure.
Non-destructive evaluations include,
inter alia, ultrasonic and acoustic
emissions, eddy currents and X-rays
(Nagarajaiah & Erazo 2016) while ground-
penetrating radar is used to assess the
condition of roads, bridge decks and piers
36
(Maser 1996; Maierhofer 2003). However,
non-destructive testing is mainly suitable
for identifying faults in the infrastructure
such as voids, or to determine pavement
thickness for example, rather than for
the assessment of the general condition
of infrastructure. Several research efforts
are focusing on the use of non-traditional
methods for condition assessment of infra-
structure. These include smart sensors,
intelligent infrastructure, machine learning
and algorithms. One such example is the
use of analytical models that have been cal-
ibrated by many assessments to determine
the condition of bridges (Aktan et al 1996).
Marcelino et al (2018), however, stressed
the cost of data collection and proposed
the development of a condition indicator
that would use a machine-learning algo-
rithm to predict pavement conditions using
less data. The use of algorithms analysing
images of corrosion taken by a robotic
system was discussed by Jahanshani and
Masri (2013). The approach improves reli-
ability of corrosion detection. The surface
condition and defects of structures can be
assessed by using fuzzy-logic algorithms to
analyse images (Pragalath et al 2018).
In new infrastructure, the latest tech-
nologies in sensors and the Internet of
Things (IoT) can be used to ensure rich data
collection that can assist in the design and
operation of infrastructure, infrastructure
condition assessment and the planning of
maintenance (Soga 2016; Aktan et al 1998;
Nagarajaiah & Erazo 2016). A number of
cities around the world, for example, use
Intelligent Transport Systems (ITS) that has
led to cost savings in the management of
the road vehicle fleet and the maintenance
of road infrastructure. This is done through
databases and interactive maps of the road
infrastructure condition depicting the type
and severity of defects (Staniek et al 2017).
Photographic and optical evaluation
has been used successfully, and improve-
ments in camera technology and the use
of Unmanned Aerial Vehicles (UAVs) such
as drones have increased the popularity
of this technology. UAVs and drones are
increasingly being used to assess the condi-
tion of infrastructure, especially in places
that are hard to reach (Ham et al 2016).
UAVs carrying cameras and thermography
cameras combined with deep-learning
technologies and neural network analysis
can be used to determine cracks in con-
crete and in asphalt (Wu et al 2018). Uddin
(2011) reviewed the use of remote sensing
satellite imagery and Light Detection and
Volume 63 Number 2 June 2021
Ranging (LIDAR) technologies combined
with Geographical Information Systems
(GIS) to enhance practices of infrastruc-
ture inventory, condition assessment and
environmental applications. The study
found that accurate assessment of pave-
ment surface distresses, condition and
maintenance quantities can be enabled
through the use of an airport GIS map and
geospatial analysis of LIDAR dense point
cloud data and intensity images.
Sewer system inspection is conducted
with Closed Circuit Television technol-
ogy which consists of a camera gener-
ally mounted on a crawler or tractor. The
inspection is conducted from manhole to
manhole, and the analysis of the images
provides details of the type and location
of defects. These include pipe cracks, joint
offsets, leaks, debris, sediment and root
intrusions. Caradot et al (2018) found that
the general condition of the sewer network
can be assessed with excellent accuracy with
the use of sewer condition evaluations.
The above literature review indicates
that infrastructure condition is often
assessed at the project (building, road)
level. Whilst modern technologies are
very useful in collecting infrastructure
performance and condition data at this
level, there is a need to analyse the condi-
tion of the whole portfolio of infrastructure
in a cost-effective way to provide input
into general policy development and high-
level budgeting. In contrast to the use of
electronic equipment at a project level,
which will be very expensive at a network
or portfolio level, this paper describes the
use of expert opinion to assess the condi-
tion of the portfolio of infrastructure in
South Africa. This was achieved through
the Infrastructure Report Cards and the
opinion survey as discussed below.
SAICE INFRASTRUCTURE
REPORT CARD SERIES
The grading of the condition of infrastruc-
ture and its presentation in a “report card”
is practised in a few countries, including
in the United States of America by the
American Society of Civil Engineers (ASCE
2017), Canada (Canadian Infrastructure
Report Card (CIRC 2016)), Australia
(Kaspura 2017) and the United Kingdom
(Living with Environmental Change initia-
tive (LWEC 2015)) and the Institution of
Civil Engineers (ICE 2017).
To date, SAICE has published three
Infrastructure Report Cards (IRCs) – in
Journal of the South African Institution of Civil Engineering
2006, 2011 and 2017. The research work Table 1 Trend in gradings from 2006 to 2017
for these was primarily conducted by the Sector Subsector 2006 2011 2017 Trend
South African Council for Scientific and
Industrial Research (CSIR), with the gradings Bulk water resources D+ D– D–

determined by the SAICE technical panels. Water Supply in major urban areas C+ C+ C+
The purpose of the IRC series has been to
Supply all other areas D– D– D–
point out to government, decision-makers
and the public at large, the importance of Major urban areas C– C– C–
Sanitation
maintenance of infrastructure and to factors All other areas E E– E
underlying its condition. Since the first free
elections in South Africa in 1994, significant Collection major urban areas C– C C

progress has been made in correcting imbal- Collection other areas D D D

ances in infrastructure provision – with Solid waste
Disposal in metros C C+ C+
more focus on infrastructure for the poorest,
disadvantaged communities. Particular Disposal in other areas D– D D–
focus has been placed on infrastructure for
National C B B
water reticulation and treatment, sanitation,
education, energy, health services and roads. Paved provincial D– D– D
However, the combination of limited resourc- Roads Paved metropolitan D– C– C–
es, public sector restructuring, inefficiency
and shortages of key skills has led to extreme Other paved municipal D– D D–
pressure on the condition of the public infra- Gravel E E
structure asset base (Wall & Rust 2017).
Airports ACSA-owned facilities B B+ B+
The IRCs grade public sector infra-
structure (water, sanitation, solid waste, Commercial ports C+ B– B–
roads, airports, ports, rail, electric- Ports
Fishing harbours C
ity, and hospitals and clinics) on the
following scale: Heavy-haul freight lines B B+ B+
QQ A: World-class
General freight lines C C+ C
QQ B: Fit for the future
QQ C: Satisfactory for now Rail Branch lines E D D–
QQ D: At risk of failure Passenger lines D+ C– D+
QQ E: Unfit for purpose
Gautrain A
In order to allow for a finer scale rating, the
operators “+” and “–” were used for ratings Eskom generation C+ C+ C+
in-between the main categories, i.e. an A–
Electricity Eskom transmission C+ B– B–
was one rating above a B+, etc.
The methodology to compile the SAICE Local distribution C– D D
Infrastructure Report Cards included: Hospitals C D+ D+
QQ Compilation of basic research reports Health care
based on desktop work by the CSIR Clinics D+ D D
QQ Arranging for the drafting of additional Public ordinary schools D+ D+
reports for selected sectors where the
Education Universities C+
CSIR itself does not have sufficient
expertise TVET colleges D+
QQ Moderation of the sector reports by Overall grade D+ C– D+
SAICE experts with additional inputs
where necessary and early results from Gradings:   = poor = neutral = above average
the survey Legend:
QQ Determination of the final gradings by Trends:   = down = neutral = up
SAICE
QQ Writing and publication of the Report
Card and its associated commentary by QQ Sanitation (major urban areas, all other QQ Airports (facilities owned by Airports
SAICE experts. areas) Company of South Africa only)
For all three Report Cards, the gradings QQ Solid waste management (waste col- QQ Ports (commercial ports only)
were conducted for the following sectors lection in major urban areas, all other QQ Rail (heavy-haul freight lines, general
and subsectors (with minor variations): areas) freight lines, branch lines, passenger
QQ Water (bulk water resources, supply in QQ Roads (national, paved provincial, paved lines, Gautrain)
major urban areas, supply for all other metropolitan, paved municipal, all QQ Electricity (Eskom generation, Eskom
areas) gravel roads) transmission, local distribution)

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 37
QQ Health care (hospitals, clinics)
QQ Education (public ordinary schools, Province of residence
universities, TVET colleges (techni-
58
cal vocational education and training
128
colleges). 28
The SAICE grading process was conducted Eastern Cape
by a number of peer-review groups, selected Free State
for their knowledge and expertise in each Gauteng
14
sector, to review the CSIR output and reach KwaZulu-Natal
consensus on the grading of the condition 16
Limpopo
of public sector infrastructure in each of the 21
Mpumalanga
sectors mentioned above, and also to pro-
21 North West
vide an overall grading for all public sector
infrastructure in the country. The process Northern Cape
252
was informed by a set of questions: Western Cape
QQ What is the condition of key elements of
South Africa’s infrastructure in public 130
(i.e. as opposed to private) ownership?
QQ How does this compare with the
Figure 1 R
 espondents’ demographics related to location
previous assessment(s)? What is the
overall trend, and what are the trends
by sectors? Age distribution
QQ What contributes to the condition and
its trends? What recommendations can 123
121
be made? (Wall & Rust 2017)
The three SAICE IRC publications to date
(SAICE 2006; 2011; 2017) provide a time
series which permits assessment of the 20–30
trend in infrastructure condition. The
31–40
overall grading of infrastructure in 2006
41–50
was a D+ grade. This improved to a C+ in
51–65
2011, mainly due to significant investment
by government in prior years in prepara- 137 > 65
tion for the Soccer World Cup in 2010. 180
However, in 2017 the grade regressed to a
D+ again. Table 1 (p 37) shows the sector-
by-sector gradings on a year-by-year basis.
This analysis indicates that, apart from 100
national transport infrastructure, the
general condition of infrastructure remains Figure 2 R
 espondents’ demographics per age group
stubbornly resistant to improvement despite
significant funding over the past ten years. membership on the condition of infra- provincial basis, the questionnaire then
In particular, the infrastructure that pro- structure in the subsectors mentioned. posed the following questions:
vides for the basic needs of the majority of The trend of the condition (becoming QQ What is your grade for the current
the people in South Africa, especially the better, unchanged or worse), as well as the infrastructure condition?
poor, is still in a bad state and is evidently reasons for the individual grading, was also QQ In your opinion, over the past five years,
deteriorating. Sectors that are especially explored. is that condition better, unchanged or
distressed include health, water, roads other The survey was undertaken in the worse?
than national roads, some railway lines, second half of 2017 – that is after the CSIR QQ What are the factors most influential in
schools and colleges. This indicates a signifi- sector reports had been completed, and your assessment?
cant inefficiency in the broader infrastruc- in parallel with, but independent of, the The respondents scored in the same
ture maintenance system. This is, of course, SAICE process of review and final grading grade categories as that used for the Score
harmful to service delivery. in preparation for the 2017 Report Card Card process, but these were converted
launch at the end of September 2017. to a five-point Lickert scale for statistical
The questionnaire firstly recorded analysis:
THE OPINION SURVEY general information of the respondent per- QQ A: World-class (on the Lickert scale = 5)
taining to location (province), age group, QQ B: Fit for the future (= 4)
Purpose and methodology gender, race, SAICE membership category, QQ C: Satisfactory for now (= 3)
The purpose of the survey was to obtain category of employer, and areas of exper- QQ D: At risk of failure (= 2)
the opinion of the broader SAICE tise. For each of the subsectors and on a QQ E: Unfit for purpose (= 1).

38 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 Respondents were invited to suggest reasons
Race distribution for the infrastructure condition. A drop-down
menu with the following options was offered:
2
QQ Funding
QQ Institutional capability
146 QQ Industry capability
QQ Maintenance
Asian QQ Technical design
Black QQ Technology
Coloured QQ Climate change
Indian QQ Pressure of use / loading
21 QQ No comment
Unspecified
Respondents were allowed to select more
White
39 than one reason per question. The data was
426
then analysed statistically to determine
31 average gradings per province and overall,
as well as to determine the paramount
reasons for infrastructure condition. The
results are discussed below.
Figure 3 R
 espondents’ demographics per race group

Response
SAICE membership 669 responses were received from SAICE
members across all nine provinces. The
35
response profile in terms of province of
residence, age distribution, race distri-
bution, membership type and current
140
employer type is shown in Figures 1 to
300 5. As expected, the largest number of
Associate respondents came from Gauteng, the
Corporate
Western Cape and KwaZulu-Natal. The
age and memberships profiles were evenly
Graduate
distributed. Two thirds of the respondents
Student
were private sector consultants.
The respondents were asked to rate the
condition of infrastructure in each of the
sectors and subsectors in which they have
195 expertise on a provincial level. Thus, the
results are available per province, as well as
for the country.
Figure 4 R
 espondents’ demographics per SAICE membership category
Survey results at national level
Appendix A (see page 46) gives the results at
Employer
the national level. Respondents considered
38 that sectors where there are significant chal-
26 87
lenges with infrastructure include water and
27 sanitation, some areas of waste collection,
district and municipal roads, gravel roads,
48 branch and passenger railway lines, health
Other
facilities and schools. In these categories the
Private sector – consult
average score was in general lower than 2.5.
Private sector – contract
Infrastructure in the best condition includes
Public sector department the Gautrain, national airports, national
Municipal roads and commercial ports. In these cat-
SOEs egories of transport infrastructure, the aver-
age score was higher than 3.3, with national
airports grading at 4.27 and Gautrain at
4.23. This is in line with the findings from
441 the desktop research and the SAICE panel
gradings given in Table 1. This data is repre-
Figure 5 R
 espondents’ demographics per employment category sented graphically in Figure 6.

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 39
Survey results at provincial level
For the first time, the SAICE Infrastructure Water: National
Score Card process now included informa- Water: Major urban
tion at provincial level through the survey. Water: Other
Figure 7 shows the overall (i.e. average for Sanitation: Major urban
all infrastructure sectors) score of infra- Sanitation: Other
structure condition per province. Waste: Collect urban
Waste: Collect other
The condition of infrastructure in the
Waste: Dispose urban
Western Province, KwaZulu-Natal and
Waste: Dispose other
Gauteng is better than in the other prov-
Roads: National
inces. North West scored particularly low. Roads: Provincial paved
The average score per province of each Roads: Metro paved
of the main categories of infrastructure is Roads: District & municipal
shown in Figures 8 and 9. Roads: Gravel
At subsector level the data also indicates Airports: National
that infrastructure in the Western Province, Ports: Commercial
Gauteng and KwaZulu-Natal is generally in Ports: Fishing
a better condition than that in other prov- Rail: Heavy-haul freight
Rail: General freight
inces. It can be seen that there are a number
Rail: Branch lines
of scores lower than 2 which indicate that
Rail: Passenger lines
this infrastructure is seen as at risk of failure
Rail: Gautrain
or no longer fit for purpose. Electricity: Eskom generation
Electricity: Eskom transmission
Reasons given for gradings Electricity: Local distribution
The reasons that respondents provided for Health: Hospitals
their answers are shown in Figure 10. From Health: Clinics
the graphs the following can be noted: Education: Schools
QQ For almost all infrastructure sectors, Education: Higher
the biggest challenge to infrastructure 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
condition is lack of maintenance, fol- Average score
lowed by institutional capacity and
funding. To some degree these three Figure 6 S ummary of country level scores per infrastructure class
aspects are interdependent.
QQ Infrastructure loading (overloading)
was also recognised as a significant 3.5
challenge.
QQ Design and technology had some influ- 3.3
ence, especially in the transport and
health sectors. 3.1
Figure 11 shows the infrastructure condi-
tion gradings by class and province. The 2.9
same pattern as in Figure 10 emerges, with
maintenance, institutional capacity, funding 2.7
Overall score

and loading being the main reasons for infra-

structure condition. The lack of maintenance 2.5
could of course be caused by a lack of institu-
tional capacity and/or funding. This pattern 2.3
is consistent over all the provinces, with a
few exceptions. There are a few anomalies 2.1
and outliers in airports, ports and rail due
to a low number of respondents in some of
1.9
the provinces. In rail infrastructure, health
infrastructure and education infrastructure
1.7
the data indicates that a lack of funding and
institutional capacity is more significant than
1.5
the lack of maintenance. The relative impor- Western KwaZulu- Gauteng Limpopo Mpuma- Eastern Northern Free North
tance to infrastructure condition of design Cape Natal langa Cape Cape State West
and technology is also evident in infrastruc- Provinces
ture classes that have been recently built or
upgraded, such as airports and ports. Figure 7 O
 verall score for the condition of infrastructure per province

40 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 Education

Health

Electricity

Rail

Ports

Airports

Roads

Waste

Sanitation

Water

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Average score per province
Eastern Cape Free State Gauteng KwaZulu-Natal Limpopo

Figure 8 I nfrastructure condition score for Eastern Cape, Free State, KwaZulu-Natal and Limpopo

Education

Health

Electricity

Rail

Ports

Airports

Roads

Waste

Sanitation

Water

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Average score per province
Mpumalanga Northern Cape North West Western Cape

Figure 9 I nfrastructure condition score for Mpumalanga, Northern Cape, North West and Western Cape

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 41
 Water Sanitation
35 35
30 30
25 25
20 20
Score

Score
15 15
10 10
5 5
0 0
Fun Inst Indus Maint Design Tech Clim Load Fun Inst Indus Maint Design Tech Clim Load
Reasons Reasons
National Urban Other Urban Other

Solid waste Roads

35 35
30 30
25 25
20 20
Score

Score
15 15
10 10
5 5
0 0
Fun Inst Indus Maint Design Tech Clim Load Fun Inst IndusMaint Design Tech Clim Load
Reasons Reasons
Waste collection: Urban Waste disposal: Urban National roads Paved metro roads Gravel roads
Waste collection: Other Waste disposal: Other Paved provincial roads Paved district roads

National airports Ports

35 35
30 30
25 25
20 20
Score

Score

15 15
10 10
5 5
0 0
Fun Inst Indus Maint Design Tech Clim Load Fun Inst Indus Maint Design Tech Clim Load
Reasons Reasons
Commercial Fishing

Rail Electricity
35 35
30 30
25 25
20 20
Score

Score

15 15
10 10
5 5
0 0
Fun Inst Indus Maint Design Tech Clim Load Fun Inst Indus Maint Design Tech Clim Load
Reasons Reasons
Heavy haul General freight Active branches Passenger lines Eskom generation Eskom transmission Local distribution

Health infrastructure Education

35 35
30 30
25 25
20 20
Score

Score

15 15
10 10
5 5
0 0
Fun Inst Indus Maint Design Tech Clim Load Fun Inst Indus Maint Design Tech Clim Load
Reasons Reasons
Hospitals Clinics Schools Higher education

Figure 10 R
 easons for infrastructure gradings by class

42 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
 Water Sanitation
35 35
30 30
25 25
20 20
Score

Score
15 15
10 10
5 5
0 0
EC FS GP KZN LIM MPU NC NW WC WAT EC FS GP KZN LIM MPU NC NW WC SAN
Provinces Provinces

Solid waste Roads

35 35
30 30
25 25
20 20
Score

Score
15 15
10 10
5 5
0 0
EC FS GP KZN LIM MPU NC NW WC WAS EC FS GP KZN LIM MPU NC NW WC ROAD
Provinces Provinces

Airports Ports
35 35
30 30
25 25
20 20
Score

Score

15 15
10 10
5 5
0 0
EC FS GP KZN LIM MPU NC NW WC AIR EC FS GP KZN LIM MPU NC NW WC PORT
Provinces Provinces

Rail Electricity
35 35
30 30
25 25
20 20
Score

Score

15 15
10 10
5 5
0 0
EC FS GP KZN LIM MPU NC NW WC RAIL EC FS GP KZN LIM MPU NC NW WC ELEC
Provinces Provinces

Health Education
35 35
30 30
25 25
20 20
Score
Score

15 15
10 10
5 5
0 0
EC FS GP KZN LIM MPU NC NW WC HEAL EC FS GP KZN LIM MPU NC NW WC EDUC
Provinces Provinces

Funding Institutional capability Industry capability Maintenance

Technical design Technology Climate change Pressure of use / loading

Figure 11 R
 easons for infrastructure scores by class and province

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 43
 For water and sanitation infrastructure Table 2 Comparison between expert reviews for the Report Card and the survey
across all the provinces, maintenance, insti- Sector Subsector IRC Survey
tutional capacity and loading are the most Bulk water resources D– D+
significant factors. This, combined with the
Water Supply in major urban areas C+ D+
generally low scores of below two or just over
Supply all other areas D– D
two for these infrastructure subsectors and in
Major urban areas C– D+
view of the current water scarcity issues, is a Sanitation
significant challenge. For solid waste, loading All other areas E D–
or overloading seems to play a significant Collection major urban areas C C–
role, which could be indicative of the space Collection other areas D D
Solid waste
problem associated with solid waste manage- Disposal in metros C+ D+
ment. The data also indicates that the relative Disposal in other areas D– D–
importance of timely maintenance in roads is National B B–
much more important than in the other sub-
Paved provincial D C–
sectors. This could be due to the sensitivity of
Roads Paved metropolitan C– C–
roads to water ingress. In the case of electric-
Other paved municipal D– D
ity infrastructure there is a variety of reasons
for the condition of the infrastructure. In Gravel E D
health infrastructure, apart from funding and Airports ACSA-owned facilities B+ B+
institutional capacity, loading seems to be Commercial ports B– C+
Ports
more important than for the other subsectors. Fishing harbours
Apart from a few outliers, industry Heavy-haul freight lines B+ C
capability does not seem to be a challenge.
General freight lines C C–
The data also indicates that climate change
Rail Branch lines D– D
has not yet impacted significantly on the
Passenger lines D+ D+
condition of infrastructure.
The comparison in Table 2 indicates Gautrain A A–

that the IRC gradings agreed reasonably Eskom generation C+ C

well with the survey results and were gen-
erally within one grade of each other.
CONCLUDING REMARKS
AND THE WAY FORWARD
Both the SAICE Infrastructure Report Card
and this survey indicated that South African
infrastructure is below par and deteriorating
further. Several factors have been identi-
fied as the main reasons for this situation.
However, government will need to act fast
to intervene and protect the infrastructure
assets worth trillions of Rand. According to
the South African National Roads Agency
(SANRAL) the replacement value of the
road infrastructure in South Africa alone is
more than R2.1 trillion (COTO 2014).
South Africa is not alone in this pre-
dicament. The American Society of Civil
Engineers (ASCE 2017) proposes a number
of solutions for improving the condition of
the US infrastructure. These include:
QQ Closing the investment gap by increas-
ing infrastructure investment from 2.5%
to 3.5% of Gross Domestic Product
QQ Increasing user-generated fees and
ensuring that such funds are not used to
off-set costs of other budget items
QQ New long-term funding programmes
with which to improve specific catego-
ries of deficient infrastructure
Electricity Eskom transmission
Local distribution
Hospitals
Health care
Clinics
Public ordinary schools
Education Universities
TVET colleges
QQ Improved leadership and planning to
ensure an overall vision for infrastruc-
ture, including full life cycle costing and
improved management tools
QQ Incentives for maintenance
QQ Identification of a pipeline of projects
attractive to private sector investment
QQ Preparing for the future by utilising new
methods, materials and technologies to
ensure more resilient and sustainable
infrastructure
QQ Considering emerging technologies for
infrastructure improvement and when
designing new infrastructure
QQ Funding for research and development
into innovative new materials, tech-
nologies and processes to modernise
and extend the life of infrastructure,
expedite repairs or replacement, and
promote cost savings.
The Canadian Infrastructure Report Card
(CIRC 2016) also emphasises improved
long-term planning and asset management.
B– C
D D+
D+ D
D D
D+ D+
C+ C+
D+
South Africa is, by contrast with the
USA and Canada, a developing nation
with an unusual history. Nevertheless, the
imperatives for the enhancement of infra-
structure are unsurprisingly familiar. The
SAICE IRC (SAICE 2017) provides similar
guidance in suggesting that South Africa
should focus on:
QQ The protection and care of existing infra-
structure to reduce the backwards slide
caused by theft, vandalism and abuse
QQ Development of institutional capability
and capacity through accelerated train-
ing and the promotion of competent,
ethical leadership
QQ Incentives for further human capital
development in civil engineering
disciplines to improve the prospects
of effective design, maintenance and
management of infrastructure
QQ Better cooperation between spheres
of government and increased private-
public sector collaboration

44 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
QQ Efforts to improve data manage-
ment, infrastructure monitoring and
evidence-based decisions, e.g. based on
smart technologies providing remote,
real-time data acquisition
QQ Reducing wastage from water lost
through physical leakage or commercial
losses, which is currently in the order of
30 to 40 percent.
At the same time, South Africa should
focus on new technologies to provide sus-
tainable, high-performance infrastructure
into the future similar to that suggested by
ASCE (2017) and based on local research
and development programmes, as well as
localisation programmes that will deliver
home-grown solutions.
From the survey it can be postulated
that government should:
QQ increase emphasis on infrastructure
maintenance significantly
QQ focus on institutional capacity, par-
ticularly in the areas and sectors where
infrastructure is in the worst condition
QQ raise additional funding for infrastruc-
ture in innovative ways.
Public infrastructure is a vital asset for
and driver of healthy socio-economic
development and welfare. It is therefore
of vital importance to manage and main-
tain this asset to maximise its impact
on the effectiveness of the economy and
service delivery.
ACKNOWLEDGEMENTS
This paper is published with the approval
of the Executive Manager CSIR Smart
Mobility Cluster. The South African
Institution of Civil Engineering is thanked
for running the survey and providing the
data to the researchers.
REFERENCES
Ahluwalia, S S 2008. A framework for efficient condition
assessment of the building infrastructure. PhD
Thesis, Ontario, Canada: Waterloo University.
Aktan, A E, Farhey, D N, Brown, D L & Dalal, V 1996.
Condition assessment for bridge management. Journal
of Infrastructure Systems, ASCE, 2(3): 116–134.
Aktan A E, Helmicki, A J & Hunt, V J 1998. Issues in
health monitoring for intelligent infrastructure.
Smart Materials and Structures, 7(5): 674–692.
ASCE (American Society of Civil Engineers) 2017.
Infrastructure Report Card 2017. Reston, VA: ASCE.
https://www.infrastructurereportcard.org.
Caradot, N et al 2018. Evaluation of uncertainties
in sewer condition assessment. Structure and
Infrastructure Engineering, 14(2): 264–273.
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
CIRC (Canadian Infrastructure Report Card) 2016.
http://www.canadainfrastructure.ca/en.
COTO (Committee of Transport Officials) 2014. South
African Road Network: Condition and Budget Needs
2014. Pretoria: Committee of Transport Officials.
Ellingwood, B R 2005. Risk-informed condition
assessment of civil infrastructure: state of practice
and research issues. Structure and Infrastructure
Engineering, 1(1): 7–18.
Frangopol, D M & Liu, M 2007. Maintenance and
management of civil infrastructure based on
condition, safety, optimization, and life-cycle
cost. Structure and Infrastructure Engineering,
3(1): 29–41.
Ham, Y, Han, K K, Lin, J J & Golparvar-Fard, M 2016.
Visual monitoring of civil infrastructure systems
via camera-equipped Unmanned Aerial Vehicles
(UAVs): A review of related works. Visualization in
Engineering, 4(1): 2–8.
ICE (Institution of Civil Engineers, UK) 2017. State of the
Nation 2017: Digital transformation. London: ICE.
Jahanshahi, M E & Masri, S F 2013. Parametric
performance evaluation of wavelet-based corrosion
detection algorithms for condition assessment of civil
infrastructure systems. Journal of Computing in Civil
Engineering, 27(4).
Kaspura, A 2017. Infrastructure in Australia. Canberra,
Australia: Institution of Engineers Australia.
Lenett, M S et al 1999. Condition assessment of
commissioned infrastructure using modal analysis
and flexibility. Proceedings, 17th International Modal
Analysis Conference, 8–11 February 1999, Gold
Coast, Australia, Vol 2, pp 1251–1259.
LWEC (Living With Environment Change initiative)
2015. Infrastructure Climate Change Impacts:
Report Card 2015. http://www.ukcip.org.uk/
lwec-infrastructure-report-card.
Maierhofer, C 2003. Nondestructive evaluation of
concrete infrastructure with ground penetrating
radar. Journal of Materials in Civil Engineering,
15(3): 287–297.
Marcelino, P, De Lurdes Antunes, M & Fortunato, E
2018. Comprehensive performance indicators for
road pavement condition assessment. Structure and
Infrastructure Engineering, 14(11): 1433–1445.
Maser, K R 1996. Condition assessment of transportation
infrastructure using ground-penetrating radar. Journal
of Infrastructure Systems, 2(2).
Nagarajaiah, S & Erazo, K 2016. Structural monitoring
and identification of civil infrastructure in the United
States. Structural Monitoring and Maintenance, 3(1).
National Planning Commission 2011. South African
Diagnostic Report. Pretoria: National Planning
Commission.
National Planning Commission 2012. National
Development Plan 2030. Pretoria: National Planning
Commission.
Perkins, P 2011. The role of economic infrastructure
in economic growth: Building on experience. Focus,
60: 24–33.
Pragalath, H, Seshathiri, S & Rathod, H 2018.
Deterioration assessment of infrastructure using
fuzzy logic and image processing algorithm. Journal
of Performance of Constructed Facilities, 32(2):1–13.
Presidential Infrastructure Coordinating Commission
2012. A Summary of the South African National
Infrastructure Plan. Pretoria: Office of the President.
SA Treasury 2017. Medium Term Expenditure
Framework. Pretoria: National Treasury.
SAICE (South African Institution of Civil Engineering)
2006. SAICE Infrastructure Report Card for South
Africa: 2006. Midrand: SAICE.
SAICE 2011. SAICE Infrastructure Report Card for
South Africa 2011. Midrand: SAICE.
SAICE 2017. SAICE 2017 Infrastructure Report Card
2017. Midrand: SAICE.
Scholz, S, Ngoli, B & Flessa, S 2015. Rapid assessment
of infrastructure of primary health care facilities:
A relevant instrument for health care systems
management, organization, structure and delivery of
healthcare. BMC Health Services Research, 15(1).
Schwab, K 2017. The Fourth Industrial Revolution.
World Economic Forum 2017. Davos, Switzerland.
Šelih, J, Kne, A, Srdić, A & Žura, M 2008. Multiple-criteria
decision support system in highway infrastructure
management. Transport, 12(4): 299–305.
Soga, K 2016. Innovation in instrumentation,
monitoring, and condition assessment of
infrastructure. In Gardoni P & LaFave, J (Eds).
Multi-hazard Approaches to Civil Infrastructure
Engineering. Cham, Switzerland: Springer.
Staniek, M, Macioszek, E & Sierpiński, G 2017. Travel
planning concept taking road infrastructure
condition into account. Proceedings, 2nd World
Congress on Civil, Structural, and Environmental
Engineering, 2–4 April 2017, Barcelona, Spain,
pp 1–5. doi: 10.11159/icte17.123.
Uddin, W 2011. Remote sensing laser and imagery
data for inventory and condition assessment of road
and airport infrastructure and GIS visualization.
International Journal of Roads and Airports,
1(1): 53–67.
Wall, K & Rust, F C 2017. Reporting the condition
of South Africa’s water sector infrastructure.
Proceedings, 40th WEDC International Conference,
24–28 July 2017, Loughborough, UK.
Woodward, J, Stoughton, K M, Begley, L & Boyd, B
2019. Evaluating water assets using water
efficiency framework for infrastructure intensive
agency. Journal of Water Resources Planning and
Management, 145(2). https://doi.org/10.1061/(ASCE)
WR.1943-5452.0001016.
Wu, W, Qurishee, M A, Owino, J, Fomunung, I,
Onyango, M & Atolagbe, B 2018. Coupling deep
learning and UAV for infrastructure condition
assessment automation. Proceedings, 2nd IEEE
International Smart Cities Conference (ISC2),
16–18 September 2018, Kansas City, MO.
Please turn over for Appendix A 
45
Appendix A: Summary of survey results at national level
Water National Major urban Other
Number of respondents 492 496 458
Average 2.51 2.53 1.98
Standard deviation 0.77 0.81 0.75
Coefficient of variation 31% 32% 38%
Sanitation Major urban Other
Number of respondents 303 290
Average 2.47 1.87
Standard deviation 0.90 0.81
Coefficient of variation 36% 43%
Solid waste Collect urban Collect other Dispose urban Dispose other
Number of respondents 76 73 77 73
Average 2.79 2.10 2.40 1.68
Standard deviation 0.84 0.84 0.75 0.78
Coefficient of variation 30% 40% 31% 46%
Roads National Province paved Metro paved Districts and municipalities Gravel
Number of respondents 653 653 642 649 610
Average 3.66 2.77 2.66 2.19 2.16
Standard deviation 0.91 0.88 0.80 0.82 0.83
Coefficient of variation 25% 32% 30% 37% 39%
Aiports National
Number of respondents 74
Average 4.27
Standard deviation 0.85
Coefficient of variation 20%
Ports Commercial Fishing
Number of respondents 39 37
Average 3.31 2.73
Standard deviation 0.80 0.65
Coefficient of variation 24% 24%
Rail Heavy-haul freight General freight Branch lines Passenger lines Gautrain
Number of respondents 76 79 77 83 52
Average 2.88 2.56 2.14 2.17 4.23
Standard deviation 1.08 0.81 0.77 0.87 1.08
Coefficient of variation 38% 32% 36% 40% 25%
Electricity Eskom generation Eskom transmission Local distribution
Number of respondents 65 66 66
Average 2.98 2.98 2.58
Standard deviation 0.96 0.81 0.88
Coefficient of variation 32% 27% 34%
Health Hospitals Clinics
Number of respondents 73 72
Average 2.16 2.19
Standard deviation 0.83 0.83
Coefficient of variation 39% 38%
Education Schools Higher education
Number of respondents 112 111
Average 2.30 3.16
Standard deviation 0.84 1.03
Coefficient of variation 36% 33%

46 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
Prediction of California TECHNICAL PAPER
Bearing Ratio and Journal of the South African
Institution of Civil Engineering
compaction characteristics ISSN 1021-2019
Vol 63 No 2, June 2021, Pages 47–56, Paper 1220

of Transvaal soils from FRANK J HAUPT (Pr Eng, Pr CPM, MSAICE)

indicator properties
started his career at the CSIR (Council for
Scientific and Industrial Research) in 1976 and
joined the consulting engineering fraternity
in 1982 as design engineer at a well-known
consultancy. He co-founded a consultancy in
1984 where he worked as managing director
F J Haupt, F Netterberg
until he sold it and retired in 2019. He
managed the civil infrastructure department of the consultancy with special
reference to roads and drainage. He obtained an MSc in Civil Engineering
A preliminary statistical analysis of 914 mostly Transvaal soils of average selected subgrade from the University of the Witwatersrand in 1980.

quality produced useful, highly statistically significant methods for the prediction of the Contact details:
maximum Proctor CBR and the Proctor compaction characteristics, as well as relationships Private Bag 7
Box 331
between soaked and unsoaked CBR and between Proctor and MAASHO compaction
Hermanus 7200
characteristics. Because of the well-known poor reproducibilty of the CBR, as well as the indicator South Africa
tests used, it is believed that the accuracy of the prediction methods is better than it appears T: +27 84 500 3111
from the large scatter of results. These methods do not replace CBR testing but supplement it by E: frnkhpt@gmail.com

possibly reducing the number of expensive tests and providing a check on gross errors.
INTRODUCTION
The California Bearing Ratio (CBR) is a
standard test used in the characterisation
of roadbed and many pavement materials,
and in some thickness design methods.
It is a relatively lengthy, expensive and
poorly reproducible test, which requires
a large amount of material. This material
may have to be transported over large
distances in some cases. These factors limit
the number of tests that can be carried
out economically.
On the other hand, soil is an inherently
variable material, and a relatively large
number of tests are required for adequate
characterisation. Indicator tests such as
Atterberg limits, linear shrinkage (LS)
and grading are relatively inexpensive and
require little material, although they also
suffer from reproducibility problems.
If it is accepted that the CBR is the bet-
ter test, and a reliable way of predicting it
from indicator tests can be found, then the
cost of pavement design and construction
control could be reduced and/or their reli-
ability increased. It is the purpose of this
paper to present methods to predict:
QQ the maximum soaked CBR at Proctor
compaction
QQ the Proctor compaction characteristics
QQ relationships between soaked and
unsoaked CBR
QQ Proctor and MAASHO compaction
characteristics.
DR FRANK NETTERBERG (PrSciNat, C Eng,
FSAICE, FICE, FSAIEG, HMAEG), who has a
BSc Hons and PhD in Engineering Geology, is
All the proposed relationships were devel- an independent researcher and specialist
oped from a database of mostly Transvaal consultant on pavement materials and
geotechnics and has over 50 years’
subgrade soils.
experience. His interests include marginal
Unless qualified, the term “soil” is used materials such as pedocretes, weathered
in this paper in its widest engineering sense, rocks, mining wastes and Kalahari sands, soluble salt damage to
and the term “local soils” to mean southern pavements and the use of saline materials and waters, soil stabilisation,
long-term road experiments, active clay and collapsing sand roadbeds,
African soils. The term modified “AASHO” and the problems of the dry macroclimate in general.
and not “AASHTO” is used because the heavy
Contact details:
laboratory compaction effort used in South 79 Charles Jackson Street
Africa is the old modified AASHO effort of Weavind Park
2 413 kJ/m3 that is somewhat less than the Pretoria 0184
South Africa
current heavy AASHTO : T180 effort of
T: +27 12 846 7051
2 695 kJ/m3 and which results in densities E: frank.netterberg@gmail.com
that are about 0.5% lower (DoT 1970). The
term “modified AASHTO” is therefore incor-
rect. The term “Transvaal” is used here for
convenience to mean the area included by the
current provinces of Gauteng, Mpumalanga,
Limpopo and North West.
PREVIOUS WORK
It is not the purpose of the paper to review
previous attempts to predict the CBR and
compaction characteristics, as this has been
done locally by Haupt (1980), Stephens (1988,
1990), Netterberg (1994) and Breytenbach
et al (2010). However, some discussions of
Kleyn’s (1955) method for the prediction
of the maximum soaked CBR at Proctor
compaction from the grading modulus
(GM) and plasticity index (PI) is relevant,
as the same CBR method was used for the
Haupt database described here. This CBR Keywords: CBR, prediction, indicator tests, subgrade, Proctor

Haupt FJ, Netterberg F. Prediction of California Bearing Ratio and compaction characteristics of Transvaal soils from indicator properties.
J. S. Afr. Inst. Civ. Eng. 2021:63(2), Art. #1220, 10 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n2a6 47
procedure is/was apparently only used in
central and southern Africa, including the
former Transvaal and Namibia, on centreline
survey samples for the determination of the
design CBR for cover requirements for new
roads. For this purpose, the Transvaal Roads
Department (TRD 1978) accepted either the
CBR at 93% of modified AASHO (MAASHO)
maximum dry density (MDD), i.e. compacted
at modified optimum moisture content and
effort according to the usual methods (DoT
1970; NITRR 1986) or the maximum Proctor
CBR. This method of determining cover
requirements was apparently later superseded
by a catalogue-design method (TRD 1994).
Kleyn offered his method as the “best
fit” to the test results on 1 134 samples
from road and airport jobs in central and
southern Africa, i.e. not as a replacement
for the CBR, but as a rough check on it
and as a supplement to possibly reducing
the number of these expensive, time- and
material-consuming tests. He warned that
it was only approximate unless corre-
lated for a particular type of material.
RESULTS
The database essentially consists of test
results carried out on 1 084 soil samples
from the former Transvaal Province, 28
from Namibia and 20 from the former
Cape Province (Haupt 1980), details of
which are summarised in Table 1.
The Transvaal soils were tested by
several different laboratories, but the others
were all tested at the National Institute for
Transport and Road Research (NITRR).
As part of an investigation of existing
roads in 1973 the former Transvaal Roads
Department (TRD) dug a total of 1 102 holes
in roads all over the Transvaal and tested the
subgrade (Burrow 1975). The results of this
survey were then combined with two much
smaller but otherwise similar surveys by the
authors in 1978 in a very dry area (southern
Namibia) (31 sites), and a continuously wet
area receiving rain all year round (George–
Storms River in the Cape) (29 sites). These
two smaller extensions to the Transvaal
survey were carried out to test and extend
the validity to very dry and very wet areas,
because the main purpose of the work was
to study moisture and suction in pavements
and to develop methods for their prediction
(Haupt 1981). This main objective was later
extended by Emery (1992). However, a pre-
liminary analysis of the results with a view
to developing a method for the prediction of
CBR was also carried out.
The number of valid test results used in
the analysis depended upon what variables
were used as predictors. The lowest number
of test results used was 1 037 as shown in
Table 1. The mean maximum Proctor CBR
of 914 former Transvaal results, for example,
was 18 with a standard deviation of 13.
The mean GM was 1.24 with a standard
deviation of 0.57. The “subgrade” samples
tested therefore represented a wide range
of materials ranging from G10 (the worst
allowable roadbed or subgrade material) to
G4 (potential base course) with a mean of
G7 (selected subgrade) according to cur-
rent practice (COLTO 1996, 1998). Of the
valid test results used 97% had CBR values
between 1 and 50, which means that the
relationships reported in this paper should
not be applied to CBR values much higher
than 50. Only 20 samples had higher CBRs.
LABORATORY TEST METHODS
The test methods used were those of the
Department of Transport (DoT 1970)
except that, following Kleyn’s (1955)
approach and normal Transvaal Roads
Department (TRD 1973, 1978) practice on
centreline and selected subgrade samples,
the specimens used for determination of

Table 1 Summary of descriptive statistics (Haupt 1980)

Number of
Mean Minimum Maximum
valid cases
Abbreviation

Parameter
Transvaal

Transvaal

Transvaal

Transvaal
Namibia

Namibia

Namibia

Namibia
Units

Cape

Cape

Cape

Cape
Total

Equilibrium moisture content EMC % 1 130 28 1 072 30 5.75 10.50 19.74 2.90 1.20 4.40 11.60 38.90 104

Grading modulus GM – 1 149 28 1 091 30 1.48 1.24 0.59 0.70 0.10 0.04 2.69 2.57 1.47

Liquid limit LL % 1 040 25 996 19 25.64 29.02 25.62 18 13 12 51 88 51

Plastic limit PL % 1 037 25 993 19 14.08 16.15 14.68 10 8 7 20 41 25

Plasticity index PI % 1 188 28 993 19 10.39 12.91 7.27 0 1 0 31 57 28

Bar linear shrinkage LS % 1 135 28 1 077 30 4.79 5.51 2.88 0 0.30 0 10 21.70 12

Sand equivalent SE – 1 113 28 1 055 30 20.25 14.38 13.63 9 2 0 55 46 70

Proctor maximum dry density MDD kg/m3 1 139 28 1 086 20 2 021.0 1 887.9 1 821.7 1 784 1 311 1 517 2 234 2 396 2 035

Proctor opt moisture content OMC % 1 249 28 1 086 20 10.46 12.88 14.0 6.90 6.10 8.40 18.50 31.20 24.10

Maximum CBR (Proctor) CBR % 1 132 28 1 084 20 29.20 20.33 30.53 4.00 0.90 3 62 120 72

% passing 425 μm sieve P425 % 1 137 28 1 079 30 51.93 63.09 89.47 10 11 58 94 99 100

% passing 75 μm sieve P075 % 1 137 28 1 079 30 28.57 35.85 55.93 7 4 13 66 91 96

Mean annual rainfall MAR mm 1 144 28 1 086 30 138.90 690.20 993.20 110.00 400 700 180 1 500 1 230

Weinert’s (1980) N-value N – 1 285 28 1 086 30 44.40 3.30 1.00 32.00 1.00 1.00 50.00 7.60 1

Thornwaite’s moisture index

I cm 1 105 28 1 086 30 –40.00 –4.79 10.00 –40 –45 10 –40 100 10
(Schulze 1958)

48 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering
the Proctor MDD and OMC were soaked Table 2 C
 orrelation coefficients (r) between maximum Proctor CBR and independent variables
for four days and the CBR at a penetration and descriptive statistics for the Haupt (1980) database for Transvaal soils
of 2.54 mm determined on each. By draw- Linear Descriptive statistics(2) Non-linear
ing a graph of soaked CBR against compac-
Parameter (r) Mean Std Dev Parameter (r)
tion moisture content the maximum CBR
at Proctor compaction was determined. CBR(1) – 18.02 13.01
The CBR obtained by this method is usu- P075 –0.69 37.24 17.55 Log (P075) –0.76
ally slightly higher than the CBR at Proctor
SE +0.69 13.76 6.06 e(SE) +0.69
MDD and was said to be more representa-
tive of subgrade conditions, to require less LSA(3) –0.67 – – – –
material and work, and to be more repro- GM +0.64 1.237 0.570 e(GM) –0.64
ducible on such materials than the usual
LLT(4) –0.64 35.74 11.34 Log (LLT) –0.72
method requiring accurate moulding at a
predetermined moisture content. The soil PIA(5) –0.63 – – – –
preparation method used for the grading, LLM(6) –0.63 – – – –
Atterberg limits and linear shrinkage was
MDD +0.63 1 879.17 171.03 MDD2 +0.65
the A-1(a) wet method. These methods
do not differ significantly from those in LST(7) –0.62 110.5 85.5 Log (LST) –0.78
more recent use (NITRR 1986). The grad- LSM(8) –0.61 – – – –
ing modulus was defined as the sum of
PM(9) –0.59 – – – –
the cumulative percentages retained on
the 2.00 mm, 425 µm and 075 µm sieves LS –0.58 5.94 3.61 Log (LS) +0.64
divided by 100 (Kleyn 1955) or, as actually PIT(10) –0.58 104.10 77.97 Log (PIT) –0.73
used to compile the database:
P425 –0.58 62.56 21.82 Log (P425) –0.58

GM = 3 ⎫⎪
P2000 + P425 + P075 ⎫
⎪(1)
⎭ ⎭
OMC –0.55 13.22 4.35 Log (OMC) –0.61
100 PI –0.52 12.96 8.18 Log (PI) –0.60

LL –0.47 29.08 11.38 Log (LL) –0.51

Where
P2000 = cumulative % < 2.00 mm PL –0.26 16.12 4.33 – –
(2000 µm) MAR –0.18 6.86 1.27 – –
P200 = cumulative % < 425 µm
I –0.13 -5.30 16.52 – –
P075 = cumulative % < 075 µm
N +0.07 3.33 1.21 – –

STATISTICAL ANALYSIS
Methods
The computer program used in these ana­
lyses was the then latest (about 1975) ver-
sion of the Statistical Package for the Social
Sciences (SPSS). However, the significance
testing was done later by hand calculator.
It is important to note that all the
predictors (e.g. PI) used in this analysis
suffer from poor reproducibility and thus
include implicit errors of measurement.
This means that classical regression theory
may not strictly be applied to these predic-
tors, because it assumes that predictors are
known accurately and with a normal distri-
bution of error. Although this assumption
was therefore not always true, there was no
readily available statistical theory to treat
uncertainties in both the predictor and the
predicted. Experience has, however, shown
that if the normal regression theory is used
in this case, the resultant error is not unac-
ceptably large (personal communication
with J J de Jager 1978, CSIR).
Notes:
1. CBR = max Proctor 2.54 mm 4. LLT = LL0.7 (P425)0.3 8. LSM = LS (P425)
CBR after four days’ soaking 5. PIA = PI0.7 (P425)0.3 9. PM = PI (P425)
2. Haupt (1980 p 338); n = 914 6. LLM = LL (P425) 10. PIT = PI (P425)0.5
3. LSA = LS0.7 (P425)0.3 7. LST = LS (P425)0.7
The basic contribution characteristics incorporated in this analysis and, although
of each of the variables to be used in the they do not strictly satisfy the assumptions
subsequent statistical analysis was first of regression, the technique has been suc-
determined. This information on the dis- cessfully applied elsewhere (personal com-
tribution and central tendencies of the vari- munication with J J de Jager 1978, CSIR).
ables provided the necessary information The structure of the data, on the other
required for the selection of subsequent hand, made it difficult to carry out a statis-
statistical techniques (Tables 1 and 2). tical test for linearity, but the plots (Haupt
A multiple stepwise linear regression 1980) clearly show that not all parameters
was performed next. This required the have linear relationships with the moisture
variables to be measured on the interval content. Various different functions were
(distances between categories are defined used to transform the variables and to
in terms of fixed and equal units) or ratio optimise the transformations.
(as interval level but with inherently A multivariate stepwise linear regres-
defined zero point) level and that the rela- sion was then carried out on the results
tionships among the variables are linear on all 914 samples. From the correlation
and additive. These restrictions are not coefficients between the relevant variables
absolute, however, and nominal variables given in Tables 2 and 3, it is apparent that
can be incorporated into the regression many of the parameters that correlate
using various techniques. They have been highly (r ≥ 0.8) with CBR also correlate

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 49
Table 3 Inter-parameter correlation coefficients (r) method. The preliminary models devel-
Parameter OMC MDD CBR SE LL PI LS P425 P075
oped by Haupt (1980) are summarised in
Table 4.
MDD –0.92 – – – – – – – – These equations show that, with only
CBR –0.55 0.65 – – – – – – – grading and LS results, the CBR can be pre-
dicted with a correlation coefficient of 0.83
SE –0.63 0.70 0.70 – – – – – –
and SEE of about 8 at a significance level
LL 0.88 –0.78 –0.47 –0.52 – – – – – better than 0.1%. These equations repre­
PI 0.95 –0.76 –0.53 –0.58 0.95 – – – – sent only a preliminary analysis of mostly
subgrade materials (i.e. relatively low CBR).
LS 0.95 –0.76 –0.54 –0.59 0.95 0.98 – – –
They significantly underestimate CBRs
P425 0.45 –0.82 –0.66 –0.67 0.22 0.30 0.28 – – above 50 and should not be indiscriminately
P075 0.75 –0.82 –0.70 –0.77 0.58 0.61 0.63 0.81 – applied to better quality materials.
It is believed that the general method
GM 0.53 0.69 0.69 0.69 0.31 0.39 –0.38 –0.98 –0.88
was sound in that a stepwise multivariate
approach of likely factors was used. The
Table 4 Summary of useful CBR prediction models transformation and adjustments of LS and
Equation
P425, etc, reported here are the optima
CBR = R α< SEE found for the prediction of equilibrium
No
moisture content rather than for CBR. It is,
2 2.1(eGM) – 23 log {LS (P425)0.7} + 54 0.83 0.001 7.6
however, likely that something other than
3 96.3 – 17.8 log {LS (P425)0.7} – 28.7 log (P075) 0.83 0.001 8.5 a simple product of, for example, LS and
4 97.7 – 17.1 log {PI (P425)0.5} – 30.7 log (P075) 0.81 0.001 8.0 P425 would also give the best correlation
with CBR. In view of the better correlation
5 119.6 – 33 log {LL0.7 (P425)0.3} – 33.2 log (P075) 0.79 0.001 8.4
coefficient for P075 of –0.69 than for P425
6 80.5 – 32.3 log {LS (P425)0.7} 0.77 0.001 8.7 of –0.58 it is recommended that combina-
7 90 – 47.4 log (P075) 0.77 0.001 8.7 tions with the former be tried.
Sufficient statistical parameters are
Notes: CBR = maximum 2.54 mm Proctor CBR after four days of soaking; x = 18.0; std dev = 13.0; n = 914
reported to calculate characteristic values
which in engineering design normally refers
highly with each other. This implies that, single parameters were the percentage to the 95% confidence interval. In this case
if one of these is used in a regression passing the 075 µm sieve (P075) and the the 85% confidence interval is reported due
equation, little improvement in accuracy sand equivalent (SE), both with correla- to the variability in the predictor values.
will result from adding another, since they tion coefficients of –0.69. It is therefore For example, in Equation 2 with GM = 2.0,
explain the same variance. recommended that research on the sand LS = 4.5 and P425 = 35 the predicted
Because the LL, PI and LS are deter- equivalent be carried out with a view to CBR (PCBR) = 29.6. The min/max char-
mined on the fraction finer than 425 µm, substituting it for the Atterberg limits. acteristic value would be = PCBR ±Z(SEE/
simple products between these and the The next best was the adjusted linear mean)*PCBR where Z = 1.04 at the 85%
P425 and more complex adjustments were shrinkage (LSA: r = –0.67) followed by confidence interval, SEE = 7.6 and mean =
determined (Table 2). The adjustments the grading modulus alone (GM: r = 0.64). 18. Therefore the minimum CBR at the 85%
shown are actually those found to correlate Of the simple products of a soil constant confidence level = 29.6 – 1.04(7.6/18)*29.6 =
best with the equilibrium moisture content and the P425 the modified LL (LLM: 16.6, rounded off to 17.
in pavements and are not necessarily those r = –0.63) was the best, followed by the Haupt’s six models were investigated by
which correlate best with CBR. These still modified LS (LSM: r = –0.61) and the Stephens (1988, 1990) on Natal soils – he
have to be found. modified PI (PM: r = –0.59). The LS was found “a very large scatter of results” and
The results are expressed in terms of the best (r = –0.58) of the common, simple rather “minimal agreement with correlated
single (r) and multiple (R) correlation coef- plasticity parameters, followed by the PI data” for all of them. Comparisons were
ficients, standard errors of estimate (SEE) (r = –0.52), with the LL being the worst apparently carried out at 100% MAASHO
and significance level (α). The square of the (r = –0.47). This confirms Kleyn’s (1955) density for CBR values of less than 50 and
correlation coefficient explains the propor- observation that all attempts to correlate at 95% for all values. These were not the
tion of the variance accounted for by the LL with CBR failed. maximum CBR (at Proctor compaction) as
particular relationship. The SEE represents Although the LL was the worst of used by Haupt but were derived graphically
the standard deviation of the actual values the LL, PI and LS, it was the best of the from specimens all compacted at MAASHO
around the prediction values and provides simple products with P425. The correla- OMC at different efforts. Plots of Haupt’s
an estimate of the prediction accuracy of tion coefficient found by Haupt (1980) for adjusted or transformed parameters LLT,
the particular regression equation. the P075 of –0.69 was slightly better than PIT and LST against CBR were said to have
that for GM (+0.64). It is recommended given similar correlation coefficients to
Prediction of CBR that Kleyn’s method should be evaluated Haupt’s for dolerites, shales and laterites
Table 2 shows that, when analysing the against Haupt’s database, which contains only. However, as the equivalent untrans-
data for linear relationships, the best the correct type of CBR results for Kleyn’s formed or unadjusted parameters gave

50 Volume 63 Number 2 June 2021 Journal of the South African Institution of Civil Engineering

100
80
Predicted maximum Proctor CBR
l=
ua
60
t
Ac
40
20
0
0 30
Actual CBR at 95% MDD AASHO compaction
Figure 1 Predictions of CBR at 95% MAASHO compaction by Kleyn’s (1955) method (Stephens 1990)
equivalent or better correlations in most
cases, further work was limited to them.
According to Stephens (1990), Kleyn’s
method predicted the CBR of Kwazulu-
Natal soils most closely at 95% MAASHO.
Stephens’ (1988, 1990) condemnation of
these and some other models appears too
summary and too severe. Figure 1, for exam-
ple, does show a large scatter and his equa-
tion for Kleyn’s (1955) method (transformed
here by the authors as Equation 8) yielded a
correlation coefficient (r) of only 0.53 when
attempting to predict the 95% MAASHO
CBR of 1 418 Natal soils, but significance
testing by the authors showed that in view
of the large number of results it is highly
significant at better than the 0.001 level.
CBR MP = 16.7 e a
Where
CBR MP = Maximum Proctor CBR
a = (12GM – PI)/18.5
As had been pointed out by Netterberg
(1994) and others, much, perhaps most, of
the scatter in such predictions is probably
due to the well-known poor reproducibility
of both the CBR and indicator tests, as well
as the different CBR method employed in
this case.
Haupt did not provide scattergrams of his
models, but the scatter in Figure 1 is prob-
ably approximately what would have been
expected from the standard error of estimate
(SEE) and accuracy given for Equation 2 in
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
ed
ic t
ed
Pr
60 90 120 150
Table 4. It can therefore be concluded that
Haupt’s models (at least Equation 2) are little
better than that of Kleyn (1955) insofar as
the accuracy of estimate at 95% MAASHO
is concerned. Figure 1 also shows that, on
average, it is overpredicting the CBR at 95%
by about 10%, and an average of 93% may be
more appropriate.
It is recommended that Haupt’s equa-
tions be evaluated against a database
containing maximum Proctor CBRs and
that various combinations of P425 and P075
with plasticity and shrinkage parameters
also be tried. In particular, the use of the
simple product of the P425 and the linear
shrinkage seems to be the simplest and best
(Stephens 1988, 1990) and has been used in
local specifications for calcretes for unsealed
(8) and sealed roads (e.g. Netterberg 1978, 1982;
COLTO 1998) to ensure a minimum CBR.
Models developed by Semmelink (1991)
for the prediction of the unsoaked CBR
immediately after compaction from the
grading, LL, LS, particle shape and fac-
tors for density and moisture content (his
Models A and F) yielded a r 2 of 0.82 and
SE of 0.22 for 1 077 “data points”. However,
none of the eleven regression constants
required was provided.
Relationship between Proctor
and MAASHO CBR
In addition to developing equations for pre-
diction of maximum Proctor CBR, models
relating CBR, MDD and OMC at Proctor
compaction with those at MAASHO
compaction, were also developed. The CBR
relationship was:
CBRP = 0.37 CBR M(9)
(r not stated; CBRM ranging from 11 to 345;
x = 88 and n = 673 local soils; α < 0.001 )
Where
CBRP = maximum 2.54 mm Proctor
soaked CBR
CBR M = maximum 2.54 mm MAASHO
soaked CBR
The average ratio of maximum MAASHO
CBR to maximum Proctor CBR of 2.7
calculated with Equation 9 is in reasonable
agreement with the ratio of 2.5 calculated
by the authors from the average factors
used by VKE Engineers (A Taute 1986
personal communication in Netterberg
1994) to convert a maximum Proctor CBR
(at an assumed 93% MAASHO) to an
ordinary (i.e. at 100 %) MAASHO CBR and
the average of 2.4 calculated by the authors
for MAASHO 93% for ferricretes from
Netterberg et al (2003) from work by Davel
(1989). For ferricretes this ratio decreased
somewhat from 2.6 at a GM of 0.5 and 1.0
to 2.3 at a GM of 2.0 and 2.5. It is therefore
recommended that both methods should
receive independent checking and that
a general method for predicting CBR at
any percentage MAASHO compaction
be developed.
Relationship between soaked
and unsoaked CBR
Although Savage (2006, 2012, 2013) pro-
posed a “one-shot” method of estimating
MAASHTO or Proctor MDD from a single
determination of compacted density and
moisture content (instead of the usual 4
or 5) and the particle bulk density and the
CBR from a single unsoaked determination
on the same specimen, his method cannot
be checked against the Haupt database
as the latter does not include particle
bulk densities. Only two local studies are
considered here, those of Haupt (1980)
and Emery (1985, 1992). Haupt’s (1980)
relationship was:
CBRUP = 1.15 CBRSP + 1.2 (10)
(r = 0.89; SEE = 5.3; n = 914; α < 0.001)
Where
CBRUP = unsoaked (at OMC) maximum
CBR at Proctor compaction
CBRSP = four day soaked maximum CBR
at Proctor compaction
51
The effect of moisture content was also eval-
uated locally by Emery (1985, 1992). He first
determined the CBR of six materials rang-
ing in quality from G1 to G10 at moisture
contents ranging from about one-quarter of
optimum to soaked. MAASHO compaction
was used for the G1-G5 and Proctor for the
G6-G10 materials. Equations relating the
soaked CBR to CBR at various ratios of mois-
ture content to OMC were then developed, as
well as one to describe the complete relation-
ship between CBR and moisture content:
CBRU = 56.54 (e–1.42 MC/OMC) CBRS0.48(11)
(r = 0.94; accuracy not stated)
Where
CBRU = unsoaked CBR at any den-
sity, at any moisture content less
than saturation compacted at
MAASHO OMC
OMC = MAASHO OMC
MC = moisture content
CBRS = soaked 2.54 mm MAASHO CBR
A similar model was developed from 245
results from the Burrow (1975) study,
converting the in-situ DCP results to labo-
ratory unsoaked CBRs at the field moisture
content, the Proctor OMCs to MAASHO
using Equation 16 and assuming the field
compaction to be equal to Proctor:
CBRU = 53.52 (e–1.12 MC/OMC) CBRS0.45(12)
(r = 0.96; accuracy not stated)
Because this equation was very similar to
Equation 11, the two sets were merged and
a general equation developed:
CBRU = 59.13 (e–1.33MC/OMC) CBRS0.46(13)
(r = 0.95; accuracy not stated)
Emery (1985, 1992) then took this as the
best model relating to soaked and unsoaked
CBR. Whilst this model appears reasonable,
in view of all the conversions and assump-
tions made it is recommended that it should
be checked against outside data, against
Black’s (1962) and Haupt’s (1980) model, and
at various densities or compactions.
Relationship between Proctor and
MAASHO compaction characteristics
Haupt’s (1980) relationship between
Proctor and MAASHO MDD was:
MDDP = 1.2 MDDM – 553 (14)
(r = 0.93; n = 783 local subgrade soils;
α < 0.001)
52
1.00
MDDP = 1.2MDDM – 553
Area enclosing about 85% of values
0.98
Proctor MDD/MAASHO MDD
0.96
0.94
0.92
0.90
0.88
1 650 1 750
Figure 2 Ratio of compactive energy versus density
Where
MDDP = Proctor MDD
MDDM = MAASHO MDD
Although it is statistically incorrect to
mathematically invert a regression equa-
tion, a similar equation to Equation 14
results if this is done to an equation
developed by Emery (1985), also from the
Burrow (1975) data file:
MDDP = 1.15 MDDM – 416 (15)
(r = 0.95 on original equation;
accuracy not stated; n = 784 local soils;
α < 0.001)
It is clear from published relationships, a
scattergram of Haupt’s (1980), and borne
out by experience, that the ratio of Proctor
MDD to MAASHO MDD is not a constant
but varies with material type and with
MDD. Ratios in the literature vary between
85 and 97% and a scattergram for 783
Transvaal soils shows a variation from 86
to 99%. Figure 2 shows the variation of
about 85% of the samples between 90–98%
with a very rough average of about 94%,
and with a tendency to increase with MDD.
This means that the ratio for granular
material is significantly higher than for
silty/clayey materials.
The optimum moisture contents are
related as follows (Haupt 1980):
OMCP = 1.25 (OMCM) – 0.5
(r = 0.92; n = 783 local subgrade soils;
α < 0.001)
Volume 63 Number 2 June 2021
1 850 1 950 2 050 2 150 2 250 2 350 2 450
MAASHO MDD (kg/m3)
Where
OMCP = Proctor OMC
OMCM = MAASHO OMC
An equation developed by Emery (1985) was:
OMCM = 0.80 (OMCP) + 0.4 (17)
(r = 0.92; n=784; α < 0.001)
Haupt (1980) also found a relation between
Proctor OMC and Proctor MDD:
OMCP = 57.5 – 0.024 (MDDP)(18)
(r = 0.92; SEE = 1.73; α < 0.001)
Such relationships as these are useful
when certain tests such as the wet-dry
brushing test must be carried out, but only
MAASHO data is available and there is a
shortage of material. It is therefore recom-
mended that they receive independent
assessment. It is also recommended that
they be evaluated on cement- and lime-
stabilised materials.
Prediction of compaction
characteristics
Only the work of Haupt (1980), Semmelink
(1991) and Allison and Webb (2002) will
briefly be considered here as they appear to
be the only local work on the subject. Haupt’s
(1980) best prediction equations were:
MDDP = 1 0.3 PI – 20.5 (LL)0.7 (P425)0.3
(16) + 2477 (19)
(r = 0.93; SEE = 66; n = 914;
α < 0.001)
Journal of the South African Institution of Civil Engineering
MDDP = 0 .83 PL – 14 (LL)0.7 (P425)0.3
+ 2364 (20) 1705 1653 1510 1415
50
(r = 0.91; SEE = 74; n = 914; α < 0.001) Maximum dry
density 1760
45
OMCP = 0.3 LL0.7 (P425)0.3 + 0.2 (PL) –1 (21) 1810
(r = 0.93; SEE = 1.59; n = 914; α < 0.001)
40 Optimum moisture
1865 content
Semmelink’s (1991) prediction equation
(his Equation 2) utilised a grading factor 35 1915

Plastic limit (%)

(GF), P425, LL and LS:
X = k 1 (GF)0.85
+ k 2(A) + k3(LS) + k4
+ k5(22)
Where
X = MDD (vibratory, in % AD), or
MDD (MAASHO, in % AD) or
OMC (MAASHO, in %)
AD = ARD × 1000
A = (P425/100)(LL/100)0.1
GF = (% passing sieve size / nominal
sieve size in mm) / 100 for sieve
sizes 75, 63, 53, 37.5, 26.5, 19, 13.2,
4.75 and 2.00 mm
Pn = nominal sieve size in mm
k n = regression constant (values not
provided by Semmelink)
Although Semmelink presented plots
showing close agreement between 21
predicted and measured MAASHO MDDs
and OMCs with r 2 of 0.90 and 0.91 respec-
tively, he did not provide the regression
constants required to use his model.
A linear regression analysis of the test
results on 5 405 samples of KwaZulu-Natal
soils, derived from a variety of geological
formations by Allinson and Webb (2002),
resulted in correlation coefficients of only
–0.24 through 0.11 to 0.46 for MAASHO
OMC against PI, PM, LL, LLM, plastic
limit (PL), GM and Semmelink’s (1991)
GF, which were mostly much lower than
those found by Haupt (1980), as shown
in Tables 2 and 3. A multiple regression
analysis using Semmelink’s (1991) Equation
2 (our Equation 22) only yielded a r = 0.44,
although all the relationships found were
highly significant. A preliminary consid-
eration according to the geological origin
yielded little or no improvement – and in
some cases was weaker. They concluded
that it was not feasible to predict accu-
rately individual OMC, MDD or CBR from
indicators, but only the average OMC and
MDD for estimation purposes.
Haupt (1980) also produced a conve-
nient chart relating Proctor compaction
characteristics to PL, LL and OMC from
Equations 20 and 21 (Figure 3). It is
A
1970
30 24 26 28 30
A3 22
2020
20
25
18
2075
16
20 14
2130 12 B
10
152190
8
6 1955 1860 1785 1590 1395 1210
2065 1675 1490 1310 1115
10
10 20 30 40 50 C 60 70 80 90 100
Adjusted liquid limit (LL)0.7 (%–0.425)0.3 (%)
Figure 3 Relationship between Proctor compaction characteristics and indicators (Haupt 1980)
recommended that these relationships G7. The average quality was therefore the
be evaluated against other data sets, and same as that of Haupt (1980 Tables 1 and
that a chart also be developed to predict 2); only the GM was slightly lower (1.09
MAASHO compaction characteristics. against 1.24). The soil origins in these two
These convenient relationships between databases are probably also broadly similar,
LLT, PL OMC and OMC enable one to although the Transvaal soils most likely
estimate any two if the other two are avail- included some calcretes and no tillites,
able by simple interpolation. Normally the whilst the Berea Red soils in KwaZulu-
cheaper Atterberg and grading parameters Natal are arguably similar to many
would be available. Table 5 shows typical Transvaal red sandy soils. The laboratory
interpolation results. test methods (except for the maximum
Proctor CBR used in the Transvaal) were
Table 5 S ummary of useful CBR prediction nominally the same. However, the soil
models preparation method may have differed, as
the KZNRD changed from the DoT (1970)
Given Interpolation
Point wet boiling method A1(a) to the A1(b) dry
PL LLT OMC MDD method in about 1970 (personal communi-
cation with R E Ashworth 2003, KZNRD).
A 34 34 16 1915
Little correlation was found between
B 17 45 16 1745 CBR and GI, GM and PI or Haupt’s (1980)
transformed or adjusted plasticity and
C 10 50 16 1675
shrinkage parameters. In fact, it was stated
Note: MDD of B is interpolated between 1915 that the simple product was as good, or
and 1675
better than the transformed or adjusted
parameters. Similar plots to that found
Evaluation against KwaZulu‑Natal soils for shrinkage modulus (LSM) were also
A later study by Stephens (1988, 1992a, found for the product of LL or PI and P425
1992b) involved the analysis of the results or P075. The correlation coefficient for
of tests on 1 421 Kwazulu-Natal soils from MAASHO CBR and LSM was found to be
the archives of the Kwazulu-Natal Roads –0.38 to –0.62 depending on the geologi-
Department (KZNRD). A wide range of cal material. However, the shrinkage was
geological material origins and engineering preferred because it was usually available
qualities was represented, ranging from even when the PI was not. No improvement
soil unsuitable as roadbed to potential was obtained with the product of the LS
base course, with an average quality of and P075. No bivariate or multivariate

Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021 53
effects were found in the relationships
between CBR and swell, modified plastic-
ity index PM = (PIx P0425) and plasticity
product PP = (PIxP075). No improvement
was obtained in plots of CBR against swell
and shrinkage modulus when they were
restricted to G5-G7 materials.
The best correlation coefficients found
were for logarithmic functions such as
log(CBR) versus log(PM) (r = +0.67 for
laterites), but simpler relationships such as
with GM alone, e.g. +0.67 for 79 laterites
and 0.62 for 133 dolerites, and P425 alone
(–0.66 for the laterites and –0.62 for the
dolerites) were almost as good. By con-
trast, for 212 sandstones, for GM alone it
was only +0.28, log(CBR) versus log(PM)
only –0.45 and none was found for P425,
illustrating the importance of geological
material as implied by Kleyn (1955).
Unfortunately, although it was recorded
and entered in the Transvaal database,
the geological origin of the material was
not considered in the preliminary analysis
reported in this paper.
DISCUSSION
From Table 2 it is clear why most of the
existing models relating CBR to other soil
parameters are non-linear. Non-linear and
logarithmic or exponential models explain
significantly more variance than linear mod-
els. It is also interesting to note that the CBR
correlates significantly better with grading
parameters (especially P075) and GM and SE
than with plasticity or compaction param-
eters, and better with the SE than the PI.
The reasons why a method developed
by one worker on a particular set of results
fails when tested by another on a different
set probably include the well-known poor
repeatability and reproducibility of both
the CBR and the indicator test methods,
the many laboratories inevitably used to
accumulate a large database, differences
in both the indicator (e.g. wet or dry and
drying temperature) and CBR (e.g. crush-
ing in all or only some oversize, scalping
at 20 or 37.5 mm) preparation and grading
(e.g. wet or dry) methods, whether, as
pointed out by Semmelink (1991), the
grading used was that before or after
compaction, the type of CBR (at OMC
or maximum, soaked or unsoaked, and
whether at 2.5 or 5.0 mm penetration) and
the problem of accurately splitting and
preparing a 200 kg sample in several bags
to the 50 g used for the plasticity tests. All
these differences are in use in southern
54
Africa and can significantly affect the
results (Pinard & Netterberg 2012).
The test methods used on Kleyn’s (1955)
South African samples were probably those
of the DoT (1948), some of which differ from
those of NITRR (1986).
Indicator tests do not of course include
factors such as the particle shape and
surface roughness, and the use of the GM,
P425 or P075, or even Semmelink’s GF
do not describe the actual grading, all of
which can be expected to affect the CBR. It
is also clear that, as Kleyn (1955) intended,
his – or as is now clear – any other method
should be calibrated on the job for the
particular materials used, which would
also only usually involve a single laboratory
and perhaps operator. Although Stephens
(1990) found a correlation coefficient
of only 0.53 for his equation for Kleyn’s
method, it was nevertheless found by the
authors to be highly significant at better
than the 0.001 level.
The most recent comprehensive
study using a large database of results
from all over South Africa and Namibia
(Breytenbach et al 2010), in which a large
number of models were tried, includ-
ing those of Kleyn (1955) and Haupt
(1980) – although apparently not that of
Semmelink (1991) – involving 28 equa-
tions for each of nearly 60 data sets,
including over 130 regressions both linear
and Weibull, failed to provide a single
accurate model. At best some could be
used only as preliminary indicators and
not as substitutes for CBR tests. Grouping
Table 6 Summary of useful relationships
Equation
Relationship
No
2 CBRP = 2.1(eGM) – 23 log {LS (P425)0.7} + 54
9 CBRP = 0.37 CBRM
10 CBRUP = 1.15 CBRSP + 1.2
14 MDDP = 1.2 MDDM – 553
16 OMCP = 1.25 (OMCM) – 0.5
18 OMCP = 57.5 – 0.024 (MDDP)
19 MDDP = 10.3 PI – 20.5 (LL)0.7 (P425)0.3 + 2477
21 OMCP = 0.3 LL0.7 (P425)0.3 + 0.2 ( PL) – 1
Where:
CBRP = Soaked CBR at Proctor compaction
CBRM = Soaked CBR at MAASHO compaction
CBRUP = Unsoaked CBR at Proctor compaction
MDDP = Max dry density at Proctor compaction
MDDM = Max dry density at MAASHO compaction
Volume 63 Number 2 June 2021
according to geological material origin
was also not effective. The “high variabil-
ity” in both the CBR and indicator results
was suggested as a possible reason for the
poor predictability.
The most accurate relationships based
on the Haupt study are summarised in
Table 6.
CONCLUSIONS
QQ Useful methods for the prediction of the
maximum Proctor CBR and the Proctor
compaction characteristics from the
grading and indicator properties were
found, as well as relationships between
soaked and unsoaked CBR, and between
Proctor and MAASHO compaction
characteristics.
QQ The high correlation coefficients are
all statistically highly significant and
indicate that the relationships found are
real and valid, at least over the range of
the database values reported.
QQ The relatively large errors of prediction
are due to the well-known poor repro-
ducibility of both the predicted CBR
and compaction characteristics, as well
as the indicator tests, the several labo-
ratories involved, and the large number
and variety of soils.
QQ The prediction error for a single labo-
ratory on a single source of material
should be better researched.
QQ As this was only a preliminary attempt,
scope probably exists for further
improvements to be made.
R
0.83
–
0.89
0.93
0.92
0.92
0.93
0.93
OMCP = Optimum moisture content at Proctor
compaction
OMCM = Optimum moisture content at MAASHO
compaction
Journal of the South African Institution of Civil Engineering
QQ Prediction methods do not replace the
CBR but supplement it by permitting
a reduction in the number of these
relatively expensive tests, by providing
a check on gross errors, and by provid-
ing a means of estimating a statisti-
cally minimum CBR. Any prediction
method should be checked on site, and
calibrated, if necessary, on the materi-
als used – as originally advised by
Kleyn (1955).
RECOMMENDATIONS
QQ More research on the sand equivalent
should be carried out with a view to
substituting it for the Atterberg limits,
as it correlated better with CBR than
the Atterberg limits.
QQ The Kleyn method should be assessed
against Haupt’s database to verify the
low correlation coefficient between CBR
and GM and PI, as reported by Stephens
(1990).
QQ The transformations of CBR with P075
should be developed, despite the fact
that the transformations of LS and P425
reported here show higher correlations,
as the latter are optima-developed for
the prediction of equilibrium moisture
content rather than CBR. P075 cor-
relates better with CBR than P425, and
such transformations may result in
more accurate prediction.
QQ Haupt’s equations should be assessed
against a database containing maximum
Proctor CBRs, and various combina-
tions of P425 and P075 with plasticity
and shrinkage parameters should also
be tried. In particular, the use of the
simple product of the P425 and the lin-
ear shrinkage seems to be the best.
QQ Emery’s (1992) best model relating
soaked and unsoaked CBR should be
assessed against outside data, against
Black’s (1962) and Haupt’s (1980)
models, and at various densities or
compactions.
QQ The relationships between OMC at vari-
ous compactive efforts reported in this
paper should be assessed independently
and extended to include cement- and
lime-stabilised materials.
QQ As the ratio of maximum MAASHO
CBR to maximum Proctor CBR report-
ed here is in reasonable agreement with
that reported by Netterberg (1994), it
is recommended that both methods
be independently assessed and that a
general method for predicting CBR at
Journal of the South African Institution of Civil Engineering Volume 63 Number 2 June 2021
any percentage MAASHO compaction
be developed.
QQ Savage’s (2006, 2012, 2013) “one-shot”
methods deserve further evaluation.
ACKNOWLEDGEMENTS
This paper is based on an MSc (Eng)
thesis submitted to the University of the
Witwatersrand by author F J Haupt under
the supervision of Prof G E Blight and Dr F
Netterberg. Both authors of this paper were
employed at the then National Institute for
Transport and Road Research of the CSIR,
Pretoria, whilst the work was undertaken.
The permission of the Director to use the
work as a thesis, and the cooperation of the
Transvaal Roads Department in making
their test results available, are acknowl-
edged with thanks.
REFERENCES
Allinson, A J & Webb, D L 2002. A study of the factors
influencing the relationship between index properties
and optimum moisture content and the formulation
of an abbreviated test method for determining
maximum dry density. Proceedings, 9th Congress of
the International Association of Engineering Geology
and the Environment, Durban, pp 168–175.
Black, W P M 1962. A method of estimating the
California Bearing Ratio of cohesive soils from
plasticity data. Geotechnique, 12(4): 271–282.
Breytenbach, I J, Paige-Green, P & Van Rooy, J L 2010.
The relationship between index testing and California
Bearing Ratio values for natural road construction
materials in South Africa. Journal of the South African
Institution of Civil Engineering, 52(2): 65–69.
Burrow, J C 1975. Investigation of existing road
pavements in the Transvaal. Report L1/75, Pretoria:
Transvaal Roads Department.
COLTO (Committee of Land Transport Officials)
1996. Draft TRH 4: Structural Design of Flexible
Pavements for Interurban and Rural Roads. Pretoria:
COLTO, Department of Transport.
COLTO 1998. Standard Specifications for Road and
Bridge Works for State Road Authorities. Yeoville:
South African Institution of Civil Engineering.
Davel, J S 1989. ‘n Statistieke analise van ferrikreet
as padboumateriaal in die voorspelling van KDV-
waardes uit die graderingsmodulus. BSc Hons. (Eng)
Thesis. Johannesburg: Rand Afrikaans University.
DoT (Department of Transport) 1948. Standard
Methods of Testing Materials and Specifications.
Pretoria: DoT.
DoT 1970. Standard Methods of Testing Materials.
Pretoria: DoT.
Emery, S J 1985. Prediction of moisture content for use
in pavement design. PhD Thesis. Johannesburg:
University of the Witwatersrand.
Emery, S J 1992. The prediction of moisture content in
untreated pavement layers and an application to
design in southern Africa. CSIR Research Report
644, Pretoria: CSIR.
Haupt, F J 1980. Moisture conditions associated with
pavements in southern Africa. MSc (Eng) Thesis.
Johannesburg: University of the Witwatersrand.
Haupt, F J 1981. Prediction of subgrade moisture
conditions. Proceedings, 10th International Conference
on Soil Mechanics and Foundation Engineering,
Stockholm, 1, Session 1, Paper 25, pp 149–156.
Kleyn, S A 1955. Possible developments in pavement
foundation design. Transactions of the South African
Institution of Civil Engineers, 5(12): 286–292.
Netterberg, F 1978. Calcrete wearing courses for
unpaved roads. Civil Engineer in South Africa,
20(6): 129–138.
Netterberg, F 1982. Behaviour of calcretes as flexible
pavement materials in southern Africa. Proceedings,
11th ARRB Conference, Melbourne, Australia,
11(3): 60–69.
Netterberg, F 1994. Prediction of CBR from indicators:
A review. Report RR90/275. Pretoria: Department of
Transport.
Netterberg, F, Davel, J S & Kruger, D 2003. Statistical
analysis of laboratory test results on South
African ferricrete as roadbuilding materials.
Proceedings, 13th Regional Conference for Africa on
Geotechnical Engineering, Marrakech, pp 591–595.
NITRR (National Institute for Transport and Road
Research, CSIR) 1986. TMH 1. Standard Methods
of Testing Road Construction Materials, 2nd ed.
Pretoria: NITRR, CSIR.
Pinard, M I & Netterberg, F 2012. Comparison of test
methods and implications on materials selection
for road construction. Proceedings , 2nd African
Community Access Programme (AFCAP 2)
Conference, 3–5 July 2012, Maputo.
Savage, P F 2006. Insights into pavement material
density and strength. Proceedings, 25th Southern
African Transport Convention, 10–13 July 2006,
Pretoria.
Savage, P F 2012. Towards a better understanding
of the moisture-density-strength relationship of
compacted soil. Proceedings, 31st Southern African
Transport Convention, 9–12 July 2012, Pretoria.
Savage, P F 2013. A rapid assessment of soil strength
parameters. Proceedings, 32nd Southern African
Transport Convention, 8–11 July 2013, Pretoria.
Schulze, B R 1958. The climate of South Africa
according to Thornthwaite’s rational classification.
South African Geography Journal, 40: 31–53.
Semmelink, C J 1991. Factors influencing the
compactability and bearing capacity of untreated
roadbuilding materials. Proceedings, 10th Regional
Conference for Africa on Soil Mechanics and
Foundation Engineering, Maseru, pp 185–192.
Stephens, D J 1988. The variation of the California
Bearing Ratio with standard classification
parameters for a selection of Natal soils.
55
 MEng Dissertation. Durban: University of
Durban-Westville.
Stephens, D J 1990. The Prediction of the California
Bearing Ratio. Civil Engineer in South Africa,
32(12): 523–527.
Stephens, D J 1992a. Variation of the California Bearing
Ratio in some synthetic clayey soils. Civil Engineer in
South Africa, 34(11): 379–380.
56
Stephens, D J 1992b. The use of a modified liquid limit
for swell prediction. Civil Engineer in South Africa,
34(12): 405–406.
TRD (Transvaal Roads Department) 1973. Manual
L1/73. Materiaalhandleiding [in Afrikaans].
Pretoria: TRD Materials Laboratory.
TRD 1978. TRD Manual L1/78. Pavement and
Materials Design Manual. Pretoria: TRD.
Volume 63 Number 2 June 2021
TRD 1994. Manual L 4/94. Materials Investigation
and Pavement Design Manual. Pretoria: TRD
Directorate Materials Branch.
Weinert, H H 1980. The Natural Road Construction
Materials of Southern Africa. Cape Town: H&R
Academica.
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 preferably be
presented in English, as the journal is distributed
internationally. Articles submitted in any of the
other official South African languages should be
accompanied by an expanded abstract in English.
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 document.
All pages should bear the authors’ names and be
numbered at the bottom of the page. With the excep-
tion of tables and figures (see below) the document
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.