Abstract
Technical Note
Estimating water pollution risks arising from road and
railway accidents
R.F. Lacey1 & J.A. Cole2
1White Cottage, Lower End, Great Milton, Oxford, OX44 7NL, UK
23 Grangefield Way, Aldwick, Bognor Regis, West Sussex, PO21 4EG, UK
both formerly with WRc plc, Medmenham, Marlow, Bucks, SL7 2HD, UK
A method is presented for estimating thefrequency of spillage of toxic liquids from roadtanker accidents. The calculation requiresinformation about the vehicular flow of tankers,
their accident rate and the probability that an accident will
result in a spill. The available sources of input data are
discussed.
Railway goods traffic, either involving liquids
transported in tanker wagons, or hazardous freight
packed in individual containers as part wagon loads, is
amenable to similar analysis but relevant input data are
much more difficult to obtain.
For any type of spill, the estimation of frequency
should be accompanied by surface and groundwater
pathway analyses, to assess the consequential damage
to the aquatic environment and any possible hazard to
public water supplies.
This paper serves as an outline of a subject calling for
more thorough study, for which improved databases,
geared to the needs of risk assessment, would be
strongly desired.
Keywords: accident, environment, pathways, railway freight, risk,
road transport, spillage, tanker, toxic hazard, water pollution
Introduction
This note outlines a procedure for quantifying the risk of
water pollution, particularly affecting surface waters and
groundwaters used as a raw water source of potable
supply, arising from accidental spillage caused by traffic
accidents on roads or railways.
The procedure was originally developed by RFL for
the road transport case, as part of a Pollution Risk
Management project, which WRc plc undertook for a
consortium of water companies in 1994. Subsequently
the procedure was extended to incorporate rail traffic
accidents.
At the time this project was begun, the authors were
not aware of any previous work in this area. Our
approach leant heavily on information contained in a
report by the Health & Safety Commission, Advisory
Committee on Dangerous Substances, 1991. Only one
more recent example is known to us of environmental
risk assessment applied to road and rail accidents,
summarized in a report by Fedra (2000). This includes a
case study of petrol and diesel fuel, transported by road
from Lisbon to Aveiras, Portugal, and a study of various
hazardous goods conveyed by rail through Alpine
valleys in Switzerland. The report cited does not go
into the detail of risk assessment, but shows how
geographical information systems, telecommunications
and Hazchem data bases could all help to cope with
accidental spills.
The question of pollution risk can be broken down
into three components:
(1) What is the risk of occurrence of a substantial spill
of pollutant along a given length of highway or
railway track?
(2) What movement, dispersal and alteration of the
pollutant will occur in its passage towards and
through surface waters (including natural and arti-
ficial drainage, streams and rivers, marshes, lakes,
reservoirs) and groundwaters (including water in
the unsaturated zone, and phreatic and confined
waters)?
(3) What hazard does the pollutant present to the
environment, to riparian water users and to water
sources used for private and public water supplies?
Here we focus mainly on component 1, but go on to
offer general advice on components 2 and 3, with
reference to relevant published work.
Explanation of terms
In everyday parlance the terms ‘hazard’ and ‘risk’ are
used imprecisely with variations of meaning. In quanti-
tative risk assessment, however, it is helpful to reduce
ambiguity by using these terms in a more standard way.
In this paper we adopt usage similar to that of the
Royal Society Study Group (Warner 1983):
A hazard is a set of circumstances (such as the transport
of a toxic liquid in a particular type of tanker over a
given distance and road category) that could lead to
a specified type of adverse event (for example a spill).
The risk of an adverse event is conveniently quantified in
terms of the frequency of occurrence (or rate of inci-
dence) of that event. Here we need to deal with fre-
quencies expressed in two different ways:
Quarterly Journal of Engineering Geology and Hydrogeology, 36, 185–192 1470-9236/03 $15.00 � 2003 Geological Society of London
f1 = accidents / million vehicle-km for a specified type of
vehicle;
f2 = accidents / km-year for a specified road and traffic
loading.
The connection between them is given by
f2 = V f1
where V is the relevant flow of traffic (millions of
vehicles per year). This relationship will be elaborated
further below.
Any spillage arising from a transport accident
becomes the hazard governing any subsequent con-
tamination of the water environment. In the reasoning
that follows we shall focus on estimating the frequencies
of potential pollution-producing events.
Cases considered
We restrict ourselves to spills arising from the com-
mercial freight of liquids in tanker lorries or rail tank
and container wagons in transit. It is necessary to
specify:
(i) the particular stretch of road or railway for which
the frequency of spills is to be estimated;
(ii) the type or types of vehicle from which spills may
occur, and the classes of initiating event;
(iii) the types of pollutant that are to be considered;
(iv) the range of spill volumes that would be of
concern.
In the following discussion we concentrate on those
types and sizes of spill that are thought to have the
most damaging further consequences, either in a single
accident or from the cumulation of accidents that are
likely to take place over a period of time.
For road transport we restrict attention to spills from
tanker lorries of over 3.5 tonnes gross vehicle weight,
caused by collision or roll-over. This would appear to
cover most large-volume spills but excludes spills from
the vehicles’ own fuel tanks or from drums or smaller
vessels carried by lorries.
For spills from rail tankers we shall concentrate on
those associated with the standard types of tank wagon
used for the transport of motor spirit. These have typical
capacity of either 32 or 75 tonnes and nominal minimum
shell thickness of 6 mm. It is important to make the
distinction between these thinner walled tanks and those
of higher specification (nominal shell thickness 11–
16 mm) used for more dangerous substances such as
liquefied gases. (Health & Safety Commission, Advisory
Committee on Dangerous Substances, 1991, appendix
8). In addition to spills from rail tankers we give some
estimates that refer to spills from freight in containers.
Our analysis excludes hazards associated with loading
and unloading. Those operations ideally should, but do
not always, take place at sites with special safeguards for
containment of spills. Sites with and without such safe-
guards merit separate risk assessments. Similarly we
have not considered the risk of accidents in areas such as
lorry parks or rail marshalling yards. It would, of
course, be important to include all such areas in a risk
audit of a water-supply catchment.
General principles
The logic and symbols applicable to estimating the
frequency of spills are explained in Figure 1. Appropri-
ate units are given in square brackets. The following
sections explain how the input quantities: V, ptanker, f1, L
and pspill can be dealt with.
For L this is straightforward since a specific length of
road or track will be considered. For pspill , the con-
ditional probability that an accident gives rise to a spill,
there is unlikely to be suitable local information and so
this must be estimated from national data. However, for
the other basic quantities, V, f1 and ptanker, decisions
have to be made whether to estimate them at a local or
national level. These decisions involve very different
considerations for the two types of transport, road
and rail, and so different approaches must be taken,
although the general principles of the final calculation
are the same.
Fig. 1. Flow diagram of spillage risk assessment.
LACEY & COLE186
Lengths of road and railway
considered
Figure 2 illustrates an imaginary area with a river and an
aquifer, traversed by a railway. Any spill in the railway
segment Xr (or indeed anywhere in the river catchment
upstream of Z) will threaten the water quality at the
pumped abstraction site Z. Any spill in the railway
segment Xa (or anywhere across the aquifer’s exposure)
will threaten the groundwater quality in the downflow
direction, which will be drawn towards the pumped
water well W. Note in this particular example the
hatched zone where both the aquifer and the river are at
risk, should a pollution incident occur there. Obviously
real situations are liable to be far more complex,
especially the interaction of surface and groundwaters,
together with the phreatic and confined zones of the
latter.
Traffic flows
Road traffic
For estimating the vehicular flow of tanker lorries on
a given stretch of road there are three possible
approaches. The choice of which to take depends on the
class of liquid to be considered and on the required
accuracy of the estimate.
Using national averages. From national traffic censuses
the Department of Transport publishes annual statistics
from which it is possible to estimate annual average
traffic flows, classified by type of road and type of
vehicle, as for example those given in Table 1, derived
from tables 3.16 and 4.9 of Transport Statistics, Great
Britain 1992 (Department of Transport, Scottish Devel-
opment Department & Welsh Office, 1993). In this table
a heavy goods vehicle (HGV) is a vehicle of over
Fig. 2. Schematic plan of a river catchment and an aquifer traversed by a railway.
Table 1. Average flows of HGVs and all motor vehicles, by class of road, 1991 data *
Road category Flow of HGVs
(million vehicles/year)
Flow of all motor vehs
(million vehicles/year)
HGVs/ All vehs.
(%)
Motorway 2.8 19.7 14.0
Built-up major roads
Trunk 0.51 6.8 7.5
Principal 0.27 5.5 4.9
Non-built-up major roads
Trunk 0.60 5.5 11.0
Principal 0.17 2.5 6.8
Minor roads 0.019 0.50 3.8
* Derived from tables 3.1 and 4.9 of Transport Statistics Great Britain 1992 (Department of Transport, Scottish Development Department & Welsh
Office 1993)
ESTIMATING WATER POLLUTION RISKS ARISING FROM ROAD AND RAILWAY ACCIDENTS 187
3.5 tonnes gross vehicle weight. Note from Table 1, that
flows of HGV not only vary by two orders of magni-
tude, as between minor road and motorways, but that
the HGV proportion increases too, in going from minor
to major highways.
Of the goods traffic in 1991 about 4% of the overall
tonnage transported involved the transport of petrol and
petroleum products, 6% involved ‘chemicals’ and 26%
involved food, drink and tobacco (Department of Trans-
port, Scottish Development Department & Welsh Office,
1993). These are the main categories of commodity to
include liquids capable of giving polluting spills. Percent-
ages such as these could be used in conjunction with Table
1 to give approximate estimates of tanker traffic for the
respective types of load. The percentages for ‘chemicals’
and for ‘food, drink and tobacco’ would, however, over-
estimate the transport of liquids because they do also
include commodities in solid and gaseous forms.
Using special surveys. If information is needed for a
specific chemical or group, or if local circumstances are
likely to be very different from the national average, there
will be no alternative to carrying out a special survey or
study. This could take the form of a tanker census on
the particular road or a study of the numbers of loads
delivered to or from key factories, stores or outlets.
Using link specific flows. There is an intermediate
approach between those outlined in the two previous
paragraphs. The Department of Transport produces
link-specific estimates of annual average daily flow for
each section of the major road network, and these can be
supplied on various media. Similar and additional infor-
mation may also be available from county highways
departments. If such estimates are not available for
HGVs but only for all motor vehicles, they could be
re-scaled for HGVs using the percentages in the final
column of Table 1. The flow of, for example, tankers
carrying petrol or petroleum products would then be
calculated by applying the commodity percentage (4%)
as in the last but one paragraph.
Rail freight traffic
While for road transport it is possible to estimate traffic
flows from readily accessible national statistics, this does
not appear to be possible for rail freight, for two reasons:
(i) From the information that is available about
specific chemicals (Health & Safety Commission,
Advisory Committee on Dangerous Substances,
1991, appendix 8) it is very clear that rail traffic is
very heterogeneous across the rail network. Route
specific information is therefore essential to
describe the types and volumes of loads conveyed.
An extreme example would be the transport of
chlorine, for which the route from Merseyside to
Anglesey carries 60% of the national traffic in that
substance.
(ii) There does not appear to be a publicly accessible
database for link-specific rail traffic loads. The
only possible sources of information would
appear to be the railway operators themselves.
Accident and spillage frequencies
For the estimation of spillage frequencies arising from
transport accidents there is no national data collection
system. For both the road and rail transport cases we are
reliant on information collected for the special studies
undertaken for or mentioned in the report of the
Health & Safety Commission, Advisory Committee on
Dangerous Substances, 1991. For spills from rail tankers
this information (from appendix 8 of the Health and
Safety Commission Report) permits only a combined
estimate of the accident involvement rate and prob-
ability of a specified spillage (see Fig. 1). For road
tankers, however, there is an advantage in carrying out a
more detailed treatment in which these quantities are
estimated separately, thus allowing for the different rates
of accidents that prevail on different classes of road.
Road accident rates
Typical accident involvement rates for HGVs and for all
road vehicles are shown in Table 2, classified by type
of road. This table was compiled from table 41 of
Road Accidents Great Britain 1991 (Department of
Transport, Scottish Development Department & Welsh
Office, 1992). There is no information specific to tanker
vehicles but it is very reasonable to assume that
their accident involvement rate would be similar to the
average for all HGVs.
The application of one of these rates to a specific
stretch of road involves the assumption that the road is
typical of its class, not much safer nor much more
dangerous. It might be thought that it would be prefer-
able to obtain accident information for the particular
road in question and hence compute a specific estimate
Table 2. Vehicle accident involvement rates, by type of road,
1991 data **
Accidents per
million vehicle-km
(HGVs)
Accidents per
million vehicle-km
(All motor vehicles)
Motorway 0.24 0.23
Built-up roads 0.88 1.52
Non built-up roads 0.52 0.60
** Derived from table 41 of Road Accidents Great Britain 1991
(Department of Transport, Scottish Development Department &
Welsh Office 1992)
LACEY & COLE188
of the accident rate. Local road-accident statistics for
each particular stretch of road are usually obtainable
from county highways departments. There are, however,
two disadvantages associated with the use of this.
(i) The available data may not distinguish tankers,
nor even perhaps HGVs.
(ii) Even if data are available for the relevant class
or classes of vehicle, the number of events may
be too small to enable a reliable estimate of
frequency to be made.
If local data are available, the recommended way of
dealing with these difficulties would be, first, to use
national rates (Table 2) to estimate the expected number
ne of accidents for the same class of vehicle and window
of time as those for which the observed number no is
obtained. Then compare no with ne. Only if no and ne are
statistically significantly different would it be worth
making adjustment for this by multiplying the relevant
rate from Table 2 by no/ne.
It is important to note that, in this procedure, no and ne
must refer to exactly the same class of vehicle and event. It
is not, however, essential that this be ‘all severity’ involve-
ments of tanker vehicles. Provided that it is reasonable to
assume that the proportional excess no/ne would be
roughly the same for all classes of vehicle, the value of this
ratio could be estimated for any class for which data is
conveniently available. The largest available class should
be chosen. We would recommend using local accident
information only after the incorporation of a local link-
specific estimate of traffic flow.
Road spillage frequencies
For road tankers the Health & Safety Commission,
Advisory Committee on Dangerous Substances, 1991
provides estimates of the average spill rate per tanker-
km, for three different types of initiating event: rollover,
collision and body material or equipment failure.
Restricting attention to traffic accidents (rollover or
collision) it is possible to use the HSC results to estimate
the probability that a traffic accident involving a tanker
will result in a spill (see Table 3). The estimates in this
table were obtained by dividing the average spill-rate per
tanker-km by the average accident involvement rate
per tanker-km. The latter was taken to be 0.656 per
million vehicle-km, an average for HGV for 1981–85,
the time period for the HSC study.
Rail accident frequencies
In the early 1960s accident statistics showed freight
derailments at about 0.2 per million train-km (i.e. 2 �
10�6 per train-km), which was comparable to the risks
of HGVs on motorways. Tanker traffic in that decade
was about 5 � 107 wagon-km/year. Some trains would
have comprised mixed loads, with only a few loaded
tankers, others would be all-tanker trains, full or empty.
To convert the wagon-km/year to train-km/year one
needs a figure for W, the average number of full tanker
wagons in all goods trains. In the absence of published
data on this, take a realistic estimate such as W = 6. The
frequency of derailment of such trains would thus be
estimated as:
(0.2� 10�6)� (5� 107)/6 per year = 1.3 per year,
involving all mixes of tanker traffic, ranging from
all-tanker cargo to trains with only one tanker-wagon.
Further information on freight trains is given by HM
Railway Inspectorate (Health and Safety Executive
1994) who reported 110 incidents of rail freight col-
lisions (12 incidents), derailments (78 incidents) and fire
(29 incidents) on the then British Railways network
during 1993/94, out of a total of 889 train accidents
affecting all classes of BR rail traffic. (The 1993/94
freight train accident statistics showed an improvement
over those for 1992/93, when freight train derailments
were twice as common). However no data appear on the
proportion of such incidents affecting full tanker wagon
loads. If, at a rough guess, that proportion were 2%, the
inference would be that 110 � 0.02 i.e. 2 train tanker
train incidents could have occurred across the whole
freight-carrying network in the year in question. This
estimate is comparable to the 1.3 tanker-train
derailments/year estimated in the previous paragraph.
The above two paragraphs are based on rather sparse
data, so are tentative. A more detailed study would
undoubtedly recognise that the risks of freight train
derailments are heavily dependent on the level of track
maintenance and also depend on the quality of rolling
stock, permitted running speeds and the complexity of
the track (with greater risks at junctions and crossovers).
Signalling error