A Risk/Cost-based Algorithm for the Routing of Dangerous Goods

Karim El-Basyouny (Corresponding Author)

Research Assistant, Dept. of Civil Engineering
University of British Columbia
Vancouver, BC, Canada V6T 1Z4
604-716-4470
basyouny@[Link]

Derek Cheng
GIS Administrator, Dept. of Civil Engineering
University of British Columbia
Vancouver, BC, Canada V6T 1Z4
604-822-8785
dcheng@[Link]

Clark Lim, [Link]., MASc.

Dept. of Civil Engineering
University of British Columbia
Vancouver, BC, Canada V6T 1Z4
604-822-8785
clim@[Link]

Tarek Sayed, [Link]., PhD.

Professor, Dept. of Civil Engineering
University of British Columbia
Vancouver, BC, Canada V6T 1Z4
604-822-4379
tsayed@[Link]

Paper prepared for presentation

at the Innovations in Traffic Control Safety Session

of the 2009 Annual Conference of the

Transportation Association of Canada
Vancouver, British Columbia
ABSTRACT

This study develops a risk/cost-based dangerous goods routing algorithm. The algorithm focuses
on mitigating the risks associated with the transportation of dangerous goods (DG) via route
selection. The algorithm was applied to a large-scale transportation network representing the
Metro Vancouver area. The network is represented spatially in a GIS database along with a real-
time dispersion plume simulating a specific chemical release under local weather conditions. GIS
facilitates the comparison between the various criteria by overlaying transportation networks
characteristics on other spatially referenced data, such as population demographics or
meteorological data. The algorithm and general methodology is used for the routing of dangerous
goods on-demand, serving individual shipments in a permitting environment. The uniqueness of
the proposed approach is in the “normalization” of risks and operating costs such that a cost-
based DG routing optimization is achieved. Furthermore, the practicality of the algorithm is
demonstrated by developing a computer application using Canadian and B.C. datasets.

2
INTRODUCTION

Each day products and materials defined as dangerous goods (DG) are shipped from one point to
another within Canada. These products include: explosives, gases, flammable liquids and solid,
oxidizing substances, poisonous and infectious substances, corrosive substances, and hazardous
waste. As such, these products and materials require special precautions to ensure their safe
transportation. The social, environmental, and monetary costs associated with DG incidents are
high and every precaution has to be taken to avoid such catastrophic events. Therefore, it is
essential to continually work towards minimizing the risk of incidents in the transportation of
dangerous goods and their potential consequences.

The Commercial Vehicle Safety & Enforcement Branch of the B.C. Ministry of Transportation is
presently collaborating with its counterpart from the Province of Alberta to develop a DG
Decision Support System (DSS) that is specific to the needs, conditions, and data availability
within B.C. The DSS should investigate the consideration of risks in the overall assessment of
optimal routing based on least-cost objectives. It is desired that the risks, and ultimately the
costs, associated with options of routing the transport of hazardous materials should consider
factors such as population centers, vehicular traffic, restrictions (i.e. height, weight, regulatory,
etc.) ecological characteristics, road classification, conditions and operations, proximity and
access of emergency services, and time of day.

This study attempts to develop a risk/cost-based dangerous goods routing algorithm (DGRA).
The algorithm focuses on mitigating the risks associated with the transportation of DG via route
selection. The algorithm is applied to a large-scale transportation network representing Metro
Vancouver area. The network is represented spatially in a GIS database along with a real-time
dispersion plume simulating a specific chemical release under local weather conditions. GIS
facilitates the comparison between the various criteria by overlaying transportation networks
characteristics on other spatially referenced data, such as population demographics or
meteorological data. The algorithm and general methodology is used for the routing of dangerous
goods on-demand, serving individual shipments in a permitting environment. This is different
than the use of such algorithms for just the planning of designated DG routes. The uniqueness of
the proposed approach is in the “normalization” of risks and operating costs such that a cost-
based DG routing optimization is achieved. Furthermore, the practicality of the algorithm is
demonstrated by developing a computer application using Canadian and B.C. datasets.

LITERATURE REVIEW

The main objective of DG routing is to determine the optimal path(s) on a network subject to
certain routing criteria. The objective, which can be either a single or multiple criteria, is
typically based on risk, equity and cost considerations. The choice of the objective criteria highly
influences the selection of the “best” route. Current literature includes different modeling
techniques to aid in the routing of dangerous goods between a given origin and destination. The
techniques differ in the type of criteria being examined and the methods by which these criteria
are combined. The most common criteria considered in the literature to route DG is risk
minimization.

3
Researchers have generally agreed that the term “risk”, in the context of DG, has to do with the
probability and the consequence of an undesirable event. Although some authors define risk as
only one of these terms (i.e., probability or consequence), it is more common to define risk as the
product of both the probability of and the consequence of the undesirable event (1,2). The
probability component is related to accident likelihood and the release probability. Undesirable
events include spills, fires or explosions in the case of flammable liquids, or a toxic clouds or
plumes in the case of pressure-liquefied gasses. The ensuing consequences include fatalities,
injuries, damages to property, losses in property values, and environmental damage. This
definition is known as "expected consequence" or as "traditional risk", primarily for the reason
that it is the definition used in the U.S. Department of Transportation (DOT) 1989 (3) guidelines
for transporting dangerous goods. Such models are probabilistic in nature and they use
conditional probabilities of accidents and the magnitude of their consequences as the two main
parameters. In general, such models differ in: (a) how the two parameters are combined to
provide a risk estimate; (b) the types and levels of detail and quality of the data acquired; and (c)
the methods of obtaining or estimating data and model parameters.

To develop probabilities some models use fault-tree analyses, whereas others use average
accident rates by mode and vehicle. To determine the magnitude of consequences, atmospheric
dispersion models and simulations are used to determine spill behavior and thus the population
exposure. Additional details on the quantification and assessment of risk and the estimation of
release rates are given in several references (4-11).

Several risk definitions and models have been proposed and examined in the literature (2,12).
Alternative risk models that include either incident probability or population exposure or both to
quantify risk were explored. The general consensus is that risk minimization problem is a bi-
criterion optimization problem; one of minimizing incident probability and population exposure.

DG routing involves a number of decisions which require the consideration of multiple and
conflicting objectives. Risk minimization seems to be the most common criterion examined in
the literature. However, other factors such as: travel time, distance, emergency responsiveness,
delay, etc. are also important. According to Leonelli et al. (13), if risk minimization is the sole
criterion for routing DG, then the routes selected are likely to be more than twice as long as the
fastest alternative and in most cases their feasibility comes in question for financial reasons. In
the literature, the models are classified according to the number of criteria included in the
analysis along with the method by which such criteria are combined to determine the best paths
between origins and destinations.

The Single Optimization Criterion (SOC) is a widely used approach to combine several criteria
into a single score or value (14-17). The new score/value is often taken to be a linear function of
different attributes such as: population exposure, distance, time and accident probability. With a
single link score/value, a simple solution method (e.g., Dijkstra’s shortest path algorithm) is used
to connect the origin-destination pair. By varying the weights of the attributes, different routes
can be generated. The process of varying the weights indicates the sensitivity that the criteria
have on route selection. Leonelli et al. (13) criticized the weighing approach since the routing
problem becomes a problem of fixing the values of the weight factors or the thresholds. In fact,
their calculations have shown that different weights (or different threshold values) can produce

4
different optimal routes. As such the determination of the optimal path becomes strongly
dependent on the decision-maker, who has to adjust the value of weights and thresholds (18).
However, this problem could be adequately addressed by having an expert panel suggest the
weights/threshold values or through the use of a weighting system (such as an analytical
hierarchy process or a genetic algorithm).

An alternative approach that has been recently gaining attention is the Multiple Objective
Criteria (MOC). This approach attempts to minimize a number of criteria simultaneously for
route selection purposes (19-21). Minimizing travel time and total population at risk is an
example application of the multiple objectives routing criteria method. Multiple criteria routing
models can be used to study tradeoffs between conflicting routing objectives. Using a MOC for
dangerous goods transportation decisions usually means that it is not possible to identify a single
“best” route but rather, the goal is to identify the set of “non-dominated” or “Pareto-optimal”
routes so that trade-offs between different objectives can be represented explicitly. However,
researchers have argued that many routes can exist within the solution set and as a result the
number of “non-dominated” paths can become very large in networks, thus rendering the
approach impractical (22). Alternatively, the Constrained Shortest Path approach is used to
minimize one attribute while limiting the sum of other attributes (22,23).

GIS MODEL

DG routing analyses should adequately integrate the road network with its surroundings, since
risk levels strongly depend on the characteristics of the region traversed by the shipments. As a
result, several researchers decided to utilize the capabilities of GIS to aid DG route planning.
Existing literature shows that the use of GIS in DG routing dates back to the early 1990’s.
Researchers have argued that the development of GIS could provide means to concurrently
analyze network topology and spatial features (18,24-26). GIS provides useful techniques for
data storage, data manipulation (for instance, to generate link attributes) and to display solutions
on a map. Moreover, GIS provides an ideal environment for design and management of DG
routes because of its ability to integrate multi-theme and multi-source data into an operational
information system. This study made use of a number of available datasets from the province of
BC. The datasets (variables) were collected and integrated into the GIS database. Information
from each dataset was used to compute the different components of the DGRA.

Road segment data for Vancouver, BC were obtained from GIS Innovations’ dataset titled
Digital Road Atlas. This dataset includes all the official designated truck routes, which are
primarily used for the transportation of DG. Spatial data including a set of EMME/2 traffic
volumes (from TransLink) and the Metro Vancouver truck routes were combined in order to
calculate the accident rates and determine route lengths.

Three types of datasets were used to describe the population distributions in the GIS database.
The first dataset contained information relating to the location of regional town centers (RTC),
which was obtained from Metro Vancouver’s dataset. The second and third datasets contained
evening and day population census information. The GIS data inventory included the Metro
Vancouver evening population distribution from the Statistics Canada 2006 Census. Calculation
of day time exposure was more complicated as the GIS data inventory did not include any day-

5
time population spatial dataset. The only data file that could be used for analysis was a table
containing 2008 day time population attribute data obtained from MapInfo. Using GIS
capabilities, the evening and day-time populations were matched. RTCs, evening and day
population census were incorporated as a second layer in the GIS environment.

An additional layer encompassed locations of emergency and special facilities. Facilities that
were most vulnerable to the impacts of DG incidents were identified as either schools or
hospitals. The number and proximity of these special facilities if an incident occurs is important.
The locations of emergency response facilities such as ambulance stations, fire stations and
police stations were necessary to determine the nearest immediate responder to commence
evacuation procedures, provide on-site traffic management, and instigate incident mitigation
measures. The relevant information about these facilities was obtained from a GIS Innovations’
dataset called Places.

DISPERSION MODEL

The exposed population is a key factor in determining the consequence of a DG release. By

overlaying the census information with the impact zone the number of individuals that might be
exposed due to a release is estimated. There are different methods to create the impact zones.
The first method uses the Emergency Response Guidebook (27) evacuation distances for large
quantity of explosives as well as isolation and protection action distances for small and large
spills. In that case, the impact distance serves as the radius that defines the impact zone. It is
possible to consider the DG shipment over a road segment as the movement of a danger circle
along that road segment. These movement carvers out a band on both sides of the road segment
thereby defining the region of possible impact.

An alternative method to estimate the impact zone is to use a computer model to effectively
incorporate climate conditions, release quantities, DG types, and topography into modeling the
release, explosion, or dispersion of DG. This method refines the determination of the impact
zone and allows for a more accurate representation of a specific chemical release. The Areal
Locations of Hazardous Atmospheres (ALOHA) dispersion model was used to simulate the
movement and dispersion of hazardous chemical gases. The tool provides estimates of pollutant
concentrations downwind from the source of a release, while accounting for both toxicological
and physical characteristics of the released material. Furthermore, ALOHA considers the
physical characteristics of the release site, atmospheric conditions and the circumstances of the
release.

This level of detail is lost when an impact zone is based on isolation and protective action
distances. These distances are based on modeling spills assuming one set of topography
conditions and quantities released. Furthermore, the guide does not indicate what constitutes a
small or a large spill and it is up to the user to determine the spill category. There are two
benefits from using a dispersion plume model. First, the model refines the estimation of the
impact zone as it is based on a number of important inputs. Second, the generated impact zone
includes both the size of the dispersion and the concentration within each level.

6
RISK/COST-BASED DANGEROUS GOODS ROUTING ALGORITHM (DGRA)

Routing of DG is a subject of considerable interest in the transportation community. There are

two major difficulties that are associated with the routing problem. First, the problem is made up
of multiple objectives (safety, security, efficiency and cost) that need to be optimized when
selecting routes, making it conceptually more difficult from the perspective of the analyst as well
as the decision-maker. Moreover, the size of the problem can become excessive (the size of the
network and the number of different DG shipments are both large). For these reasons, an initial
screening of the network should always be performed (REF).

Figure 1 shows a conceptual framework for the proposed risk/cost-based DGRA. The model
contains two major components: a Dispersion Model and a GIS. The dispersion model uses site
data, chemical, atmosphere and source information to generate a plume footprint. Afterwards, the
footprint and a number of datasets are integrated together in a GIS environment to perform the
analysis. Based on the study objectives, the proposed methodology involves a set of dangerous
goods routing criteria pertaining to safety, efficiency, security, and cost. These are chosen to
include: incident probabilities, incident consequence, operating costs, and human health impacts.

Incident probabilities were developed from B.C.-calibrated accident prediction models utilizing
traffic volumes as measures of exposure as well as segment length. Incident’s consequence is
proportional to the number of individuals exposed due to the occurrence of an incident. To
determine the incident’s consequence a dispersion model is used to effectively incorporate
climate conditions, release quantities, dangerous goods types, and topography into modeling the
release, explosion, or dispersion of DG. As a result, a chemical release is simulated using a
computer model to generate a dispersion plume footprint.

The generated footprint represents the extent of the chemical dispersion and is assumed to be of
the same concentration (the Immediately Dangerous to Life and Health concentration), with its
shape representing the spread of the released gas cloud to the level of concern. A typical
footprint diagram contains four shaded areas representing three ground level concentrations and
a 95% confidence interval. The inner most area inside the footprint is the region predicted to
have ground level concentrations above the limit that is specified during the model run. The
outer lines drawn on either side of the footprint reflect uncertainty in the wind direction and are
computed based on a 95% confidence factor.

The above information is integrated into the GIS database. By overlaying census population
information with the impact zone, the number of individuals that may be exposed (and degree of
exposure) due to a release is estimated. This was achieved by first transferring the footprints co-
ordinates from the dispersion model into the GIS database. For each concentration level a
composite footprint was generated to determine the impact zone along an entire route. The
analytical capability of the GIS software was then used to enumerate the effected population.

Finally, an optimal routing algorithm combining operating costs and risk costs was applied in the
GIS environment to allow for the evaluation of a specific truck route or alternate truck routes
between a particular origin and destination. The algorithm and general methodology were used
for the routing of dangerous goods on-demand, serving individual shipments in a permitting

7
environment. This is different than the use of such algorithms for just the planning of designated
DG routes.

The proposed routing algorithm expands on the traditional risk definition. Total cost is expressed
as a function of the traditional risk (converted into a dollar amount) plus operating costs.
Traditional risk is expressed as the expected number of individuals that are affected in case of an
incident. This is achieved by multiplying incident probabilities with consequences. Shortreed et
al. (28) suggested a 5% conditional probability for the occurrence of an incident given a traffic
accident involving a DG truck. To associate a dollar amount with this figure a cost value is
assigned for each of the expected exposed individuals. Therefore, total cost along an entire route
is considered a function of an aggregated accident probability along the route, the risks to human
health from being exposed to a certain concentration level expressed in a dollar value and, the
operating costs associated with that route. The total cost formulation is expressed as:

(2)

where
TCr Total cost on route r
ARi Accident rate (accidents per vehicle-km) for route segment i
Li Length of route segment i in (km)
m Number of concentration levels (in this study m = 4 levels)
POPj Exposed population along entire route r at a specific concentration level j
HIj Health impacts in terms of $/person due to the exposure to concentration level j
TTr Travel time for route r
OC Operating costs in terms of $/unit time

For each chemical, a footprint containing four shaded areas representing three ground level
concentrations and a 95% confidence interval was generated. A dollar amount was associated to
each concentration level. The inner most area inside the footprint is the region predicted to have
ground level concentrations above the tolerable limit. Individuals living within this region are
expected to suffer from severe health impacts. As a result, a higher dollar value was associated
with high concentration levels and vice versa. It was assumed that health costs associated with
the three levels of concentrations (from highest to lowest) are $1000, $500 and $100 per
individual. As a safety precaution, individuals residing within the 95% confidence interval (the
region representing the uncertainty in the wind direction) are evacuated at a cost $200/person.

In this study, the operating costs were assumed to be a fixed dollar amount regardless of the
shipment type, size or time of day. The operating costs were based on the carrier’s value of time
as proposed by Waters et al. (29) and were assumed fixed regardless of shipment type, size or
time of day. After conversion from US to Canadian dollars, it was estimated at CAD$55 per
hour.

SCENARIO DEVELOPMENT

For demonstration purposes three routes connecting a pair of origin and destination (O-D) were
investigated. It is assumed that only one shipment is going to traverse the O-D pair. Figure 2
depicts the study area and the three proposed routes. The three routes pass through almost the

8
same number of intersections, albeit having variable lengths. Only route (R2) passes through
Langley’s regional town centre (RTC). There are no tunnels, bridges or HOV lane present along
the three routes. Route (R1) is a major highway in the lower mainland region. The highway is
known as Highway (1) and begins at Horseshoe Bay ferry terminal in West Vancouver and
continues for 170 (km). Route (R2) traces part of Fraser highway but passes through Langley’s
town center. Route (R3) passes through a mixture of local arterials in the city of Langley.

To test the system, a hypothetical scenario involving the transport and the subsequent release of
chlorine was investigated. In order to determine the impact zone a dispersion model was used to
simulate the release. Input information is assumed to generate a footprint due to an accidental
release of a shipment of chlorine in the atmosphere. Chlorine is a nonflammable gas in liquid
state. However, like oxygen it is capable of supporting the combustion of certain substances.
Many organic chemicals react readily with chlorine, in some cases with explosive violence. To
generate the footprint, assumptions pertaining to the local weather conditions, type and amount
of chemical released were made.

Figure 3 shows the generated footprint at three concentration levels as well as the 95%
confidence interval. The footprint contains three shaded areas representing three ground level
concentrations. Using a specific import feature (in GIS) the generated footprint was transferred
from ALOHA into the GIS database. With the aid of the import feature the user is able to view
and query the footprint output with other data layers available in the GIS database. However, the
imported footprint resulted from a single release at a certain location. To determine the impact
zone along an entire route a composite plume was generated by creating multiple footprints
every 100 (m) and combining them into an impact zone as shown in Figures 4.

RESULTS

Table 1 provides a summary of the individual components of the DG routing algorithm. The
results indicate that route (R1) minimizes travel time and consequently transport cost. On the
other hand, route (R2) minimizes distance traveled as well as accident probability. Route (R3) is
probably the worst choice since it has the highest accident probability, travel distance, travel
time, and transport costs.

Table 2 summarizes population exposure numbers using the dispersion model to determine the
impact zone. The results indicate that route (R2) minimizes population exposure at the three
levels of concentrations during evening. However, during day route (R3) minimizes population
exposure at highest concentration zone as well as the number of overall evacuees.

Table 3 summarizes the number of emergency and special facilities along the three routes. Route
(R2) has the highest number of emergency response facilities among the three routes at 20
(ppm). This is very important since having several emergency response facilities along the route
will result in a more responsive reaction to the incident and facilitate isolation and protection
procedures. It is also important to note that exposure to chlorine at this concentration level is
expected to have a very severe impact on human health.

9
Furthermore, routes (R2) and (R3) pass by the least number of schools and hospitals within the
highest concentration zone. As the concentration level drops to 0.5 (ppm) route (R2) still has the
least number of exposed special facilities. This is important since it minimizes the number
evacuees in such sensitive facilities. On the other hand, route (R3) exposes 30 special facilities to
probable danger. Should an incident occur during the transportation process this route places
some school-aged children and potential ill individuals at the risk of being exposed to harmful
materials.

The DGRA combines the risks and costs together in a single routing criterion by normalizing the
risks into a cost-based approach. Table 4 summarizes the costs associated with routing DG based
on evening- and day population exposure.

For evening and day-time exposure, route (R2) minimizes the total costs. However, it could be
argued that route selection should consider not only the costs but also the number of emergency
and sensitive facilities on-route. If that is the case then route R2 is still the best choice since it
has more emergency response facilities and less schools and hospitals on route (Table 3). If a
release does occur in the evenings, the number of schools along the route is not going to be an
issue as they are typically unoccupied during this time. During the day, schools and hospitals are
expected to be operating and therefore their presence will impact the analysis. Evacuating young
children and potentially ill-individuals is a complex and time consuming process which needs to
be minimized as much as possible. In contrast, route R3 has the highest cost because: (i) it
exposes a large number of individuals to risk; and (ii) it has the highest operating cost.

DISCUSSION

Table 5 provides a summary of the “best route” by the different DGRA components. The table is
going to be referenced throughout this section to compare and differentiate between the findings
of each routing criteria. The results in Table 5 indicate that route (R2) minimized travel distance
but did not minimize travel time. The difference in the results is explained by examining the
speed profiles on both routes. The difference in speed profiles between routes (R1) and (R2) is
explained by the fact the route (R1) is a major highway in the lower mainland region. The
highway is known as Highway (1). As a result of having a higher speed, route (R1) minimized
travel time. In contrast, route (R2) passes through the city of Langley’s town centre where lower
speeds are enforced. There is an apparent similarity between the travel time and transport cost
criteria. These similarities can be explained by recognizing that transport cost is a function of
travel time and operating costs. Therefore, it is expected that both measures would identify the
same route. As expected incident probability and travel distance criteria identified the same route
since both measures are based on road segment length.

Table 6 shows the percentage increase from the base values for each criterion. For travel
distance, by choosing routes (R1) and (R3) over route (R2) the decision-maker is increasing the
distance traveled by 22% and 52%, respectively. These percentages show that for this specific
criterion, a significant difference in the results is apparent between the three alternative routes.
Routing by travel time or cost increases the percentage by 9% and 78% for routes (R2) and (R3),
respectively, relative to route (R1). This indicates that routes (R1) and (R2) could be considered
as surrogate routes since the results are not significantly different under both criteria. Similarly,

10
the incident probability criterion shows no significant difference between routes (R1) and (R2).
Route (R3) is probably the worst choice since it has the largest distance, longest time, highest
cost and incident probability while showing a significant increase in all criteria when compared
to the best alternative.

Route evaluation is generally time-of-day dependent because many of the criteria for routing and
risk assessment depend on traffic volumes and activity patterns that vary throughout the day. The
time of day is factored into the analysis by accounting for variation in population distribution
during the evening and day times. To estimate the incident consequence it is necessary to
determine the population exposure due to a release. It is expected that the number of individuals
exposed would vary by time of day. Residential neighborhoods are expected to be densely
populated during the evening whereas commercial and industrial areas are expected to be
sparsely populated. The relationship is reversed in the daytime where commercial and industrial
areas are expected to be occupied. To capture the difference in population distribution and to
properly estimate the incident consequence, the time of day was included in the analysis.
Evening and day-time population distributions for the Metro Vancouver area were integrated
with other spatially referenced data into the GIS database.

Examining the results in Table 5 reveals that route (R2) minimizes incident consequence during
the evening and route (R3) during the day. This is a reasonable finding since route (R2) passes
through Langley’s town center. The RTC is surrounded by a large number of commercial and
industrial facilities which are expected to be occupied in the day and vacant at evening.
Conversely, the landuse around route (R3) is mostly made up of residential areas that are densely
populated during the evening and sparsely populated in the day.

The results of incident probability and consequence are sometimes conflicting as seen in Table 5.
Route (R2) minimizes incident probability while routes (R2) and (R3) minimize incident
consequence during the evening and daytime, respectively. In general, the traditional risk
criterion is adequate as it accounts for both probability and consequence, but it fails to account
for the operating and health impacts due to a DG release. As a result, the DGRA was adopted to
expand the traditional risk definition. The approach normalizes the costs and risks into one
routing criteria based on probability, consequence, human health impacts and operating costs.

The DGRA combines both the incident probability and consequence to estimate the risks
associated at each concentration level. Exposure to different levels of chlorine concentration
results in a variety of health impacts. Since using a dispersion model allows for the determination
of the concentration levels within each impact zone then a cost value can be associated with each
level. In essence, the approach recognizes that not all individuals are going to be affected in the
same manner and that severe health impacts can result due to exposure to high levels of
concentration. The approach calculates the indirect costs due to the occurrence of an incident
involving DG. By changing the risks into a dollar value the indirect costs can be summed to
other costs such as operating cost. Route (R2) minimized the overall costs for evening and
daytime exposure.

11
There are two important findings to note: (i) the indirect costs associated with each of the three
routes exceeded the operating costs by a large margin; and (ii) the indirect costs are higher
during the day than during the evening.

CONCLUSIONS & FUTURE RESEARCH

In this study a risk/cost-based DGRA was developed. The algorithm had four major components.
First, a chemical release was simulated using a dispersion model to determine the impact zone
(IZ). The generated IZ is based on time of day, weather conditions and type/amount of chemical
released. Second, the DGRA was formulated to include a set of routing criteria including:
incident probability, population exposure, travel distance travel time and health cost. Third, all
the above information was integrated into a GIS database. Lastly, the GIS database utilized an
optimal routing algorithm to allow for the evaluation of a specific truck route or alternative truck
routes between a particular origin and destination. The applicability of the algorithm was
demonstrated using B.C. calibrated datasets representing the Metro Vancouver area.

There are many other uses of GIS models as DG incident management tools to identify
evacuation plans, emergency responsiveness, re-routing of existing traffic, etc. The current study
could be extended to include additional datasets. For example, locations of farms, crops,
watersheds and other environmental locations can be used to determine the effects on the
environment. The number of students enrolled in schools can be obtained to determine the exact
number of evacuees; with a similar count for hospitals. Moreover, information on bus-stops,
shared bicycle or bus lanes, transit stations, etc., can be factored into the analysis. Other topics
include determining adequate release probabilities, obtaining accurate and recent truck accident
rates, using operating speeds, and allowing some of these variables to vary by time-of-day.
Finally, the health costs associated with an exposure to certain DG needs further investigation.

REFERENCES

1. Covellov , T. and Merkhofer, M. W. (1993) Risk Assessment Methods: Approaches for

Assessing Health and Environmental Risks. Plenum Press, NY.
2. Erkut, E. and Verter, V. (1998) Modeling of Transport Risk for Hazardous Materials.
Operational Research Letters. 46, 625-642.
3. USDOT (U.S. Department of Transportation) (1989). Guidelines for Applying Criteria to
Designate Routes for Transporting Hazardous Materials. Report No. DOT/RSPA/OHMT-
89-02, Federal Highway Administration, Washington, DC.
4. Abkowitz, M., A. Eiger and S. Srinivasan (1984). Estimating the Release Rates and Costs of
Transporting Hazardous Waste. Transportation Research Record 977, pp. 22-30.
5. Pijawka, K., S. Foote and A. Soesilo (1985). “Risk Assessment of Transporting Hazardous
Material: Route Analysis and Hazard Management,” Transportation Research Record 1020,
pp. 1-6.
6. Radwan, E., K. Pijawka, A. Soesilo and F. Shieh (1986). “Transportation of Hazardous
Wastes in Arizona: Development of a Data Base Management System for Basic Analysis,”
Transportation Research Record 1063, pp. 1-7.

12
7. Abkowitz, M. and Cheng, P.D.-M. (1988) Developing a risk/cost framework for routing
truck movements of hazardous materials. Accident Analysis & Prevention, Vol. 20, No. 1,
pp. 39-51.
8. Glickman, Theodore S. (1991). “An Expeditious Risk Assessment of the Highway
Transportation of Flammable Liquids in Bulk,” Transportation Science 25, pp. 115- 123.
9. Harwood W.D., Viner J.G, and Russell, E.R. (1991) Truck Accident and Release Rates for
Hazmat Routing. ASCE Conference Proceeding Paper, pp. 177-190.
10. Harwood W.D., Viner J.G, and Russell, E.R. (1993) Procedure for Developing Truck
Accident and Release Rates for Hazmat Routing. J. Transp. Eng. 119, 189-199.
11. Saccomanno, F. and J. Shortreed (1993). “HAZMAT Transport Risks: Societal and
Individual Perspectives,” Journal of Transportation Engineering 119, 177-188.
12. Erkut, E. and Ingolfsson, A. (2004) Transport Risk Models for Hazardous Materials:
Revisited. Operational Research Letters. 33, 81-89.
13. Leonelli, P., Bonvicini, S. and Spadoni G. (2000) Hazardous Materials Transportation: A
Risk Analysis based Routing Methodology. Journal of Hazardous Materials, 71, 283-300.
14. Huang, B., Cheu, R.L. and Liew, Y.S. (2003) GIS-AHP Model for HAZMAT Routing with
Security Considerations. IEEE 6th International Conference on Intelligent Transportation
Systems, Shanghai, China.
15. Huang, B., Cheu, R.L. and Liew, Y.S. GIS (2004) GIS and Genetic Algorithms for
HAZMAT Route Planning with Security Considerations. International Journal Geographical
Information Science. VOL. 18, NO. 8, 769–787.
16. Dilgir, R., Zein, S.R. and Al Popoff (2005) Dangerous Goods Route Selection Criteria. Paper
prepared for presentation at the Towards a Safer Road Network Session of the 2005 Annual
Conference of the Transportation Association of Canada Calgary, Alberta.
17. Han, K. (2006). Developing a GIS-based Decision Support System for Transportation
System Planning. Paper submitted to AASHTO GIS-T 2006 Student Paper Contest.
18. Panwhar S.T., Pitt, Robert. and Anderson, M.D. (2000) Development of a GIS-Based
Hazardous Materials Transportation Management System. Publication of the University
Transportation Center for Alabama. Report, 99244.
19. Nema, A. K. and Gupta, S. K. (2003) Multiobjective Risk Analysis and Optimization of
Regional Hazardous Waste Management System. Practice Periodical of Hazardous, Toxic,
and Radioactive Waste Management. ASCE, 69-77.
20. Huang, B. and Fery, P. (2005) Multiobjective Optimization for Hazardous Materials
Transportation. Transportation Research Record: Journal of the Transportation Research
Board, pp 64-73
21. Zhang, M., Ma, Y, and Wen, K. (2005). Location-routing model of hazardous materials
distribution system based on risk bottleneck," ICSSSM, pp.362-368 Vol. 1, Proceedings of
ICSSSM '05. 2005 International Conference on Services Systems and Services Management,
Vol. 1
22. Frank, W.C., Thill, J.-C. and Batta, R. (2000) Spatial decision support system for hazardous
material truck routing. Transportation Research Part C, Vol. 8, pp. 337-359.
23. Jin, H., Batta, R., and Karwan, M.H. (1995). On the Analysis of Two New Models for
Transporting Hazardous Materials Operations Research, Vol. 44, No. 5, pp. 710-723
24. Lepofsky, M., Abkowitz, M. and P. Cheng (1993) Transportation Hazard Analysis in
Integrated GIS Environment. Journal of Transportation Engineering, Vol. 119, No. 2, pp.
239-254.

13
25. Brainard, J., Lovett, A., and ParfittT, J. (1996) Assessing hazardous waste transport risks
using a GIS. International Journal of Geographical Information Systems, 10, 831–849.
26. Lovett A, Parfitt JP, and Brainard, JS. (1997) Using GIS in risk analysis: A case study of
hazardous waste transport. Risk Anal 17(5):625-33.
27. Emergency Response Guidebook (2008) US Department of Transportation, Transport
Canada, Secretariat of Transport and Communications of Mexico.
28. Shortreed, J., Belluz, D., Saccomanno, F.F., Nassar, S., Craig, L., and Paoli, G. (1994).
Transportation risk assessment for the Alberta special waste management system (Final
Report). Waterloo, Ontario, Canada: Institute for Risk Research, University of Waterloo.
29. Waters, W. G., Wong, C. & Megale, K. (1995) Value of Commercial Vehicle Time Savings
for The Evaluation of Highway Investment A Resource Saving Approach Journal of Public
Economics, 35

14
Table 1: Risk Measures
Routes
Description
R1 R2 R3
Accident Probability 4.10E-04 3.95E-04 5.88E-04
Travel Distance (km) 33 27 41
Travel Time (min) 23 25 41
Transport Cost ($) 21 23 38

Table 2: Exposed Population by Time of Day

Routes
Description
R1 R2 R3
20 (ppm) 41,163 36,230 41,289
Evening Exposed 2 (ppm) 19,932 16,206 21,085
Population 0.5 (ppm) 9,503 4,538 14,339
Evacuated 92,870 80,942 90,315
20 (ppm) 63,584 63,581 47,608
Day Exposed 2 (ppm) 19,861 9,768 13,194
Population 0.5 (ppm) 6,421 2,756 15,622
Evacuated 104,841 104,632 85,866

Table 3: Number of Special & Emergency Facilities alongside each Route

Concentration No. of facilities on each route
Facility Type
Level R1 R2 R3
Sensitive 11 9 9
20 ppm
Emergency 1 3 2
Sensitive 20 21 18
2 ppm
Emergency 4 5 5
Sensitive 25 24 30
0.5 ppm
Emergency 5 7 8

Table 4: Costs Associated with routing DG during Evening

Evening Day
Description
R1 R2 R3 R1 R2 R3
20 (ppm) 16,858 14,311 24,259 26,040 25,114 27,972
Risk
2 (ppm) 4,082 3,201 6,194 4,067 1,929 3,876
Costs
0.5 (ppm) 389 179 842 263 109 918
($)
Evacuated 7,607 6,394 10,613 8,587 8,266 10,090
Total 28,936 24,085 41,909 38,958 35,418 42,856
Operating Costs ($) 21 23 38 21 23 38
Total Cost ($) 28,957 24,108 41,947 38,979 35,441 42,894

15
Table 5: Summary of the Routing Criteria Results
Routes
Description Time
R1 R2 R3
Travel Distance All √
Travel Time All √
Transport Cost All √
Incident Probability All √
Evening √
Incident Consequence
Day √* √
Evening √
Risk/Cost DGRA
Day √
* Based on concentration levels 2 (ppm) & 0.5 (ppm)

Table 6: Percentage Increase from the base Value

Routes
Description
R1 R2 R3
Travel Distance (km) 22% Base 52%
Travel Time (min)
Base 9% 78%
Transport Cost ($)
Incident Probability (per million) 4% 49%
Evening 20% Base 74%
Total Cost ($)
Day 10% 21%

16
Figure 1: Dangerous Goods Routing Algorithm: Conceptual Framework

Figure 2: Study Area and Evaluated Routes

17
Figure 3: ALOHA Generated Plume Footprint for Chlorine

Figure 4: GIS Generated Impact Zones for all three Routes

18