December 1996 € Volume 23 € Number 6

The Road Not Taken

Decision support systems provide sound directions for transportation of hazardous materials

By Erhan Erkut

SUPPOSE YOU WISH TO TRAVEL from Edmonton, Alberta, to Banff, Alberta, for a weekend getaway in the family van. Which route would you take? This cannot be too hard; after all, there are only two main routes (with minor detour options) to choose from. If you want the fastest route, you would choose to take Highway 2. On the other hand, if you like scenic routes, you may prefer driving through Jasper. Yet, there is a higher risk of colliding with a mountain sheep or an elk on the scenic route. On the other hand, there is probably a higher risk of getting a speeding ticket on Highway 2, since you would probably drive slower on the scenic route to soak in the beauty of the mountains and lakes, as well as to avoid collisions with animals. On the other hand (you probably know already that academics never run out of hands), Highway 2 offers more bathroom and junk food stops for the kids in the van.

The route-selection decision may not be that easy after all; it depends on what you are willing to consider as relevant to the decision. (Of course, if you like variety, you may take one route to go to Banff and the other one to get back to Edmonton.)

Now consider the more difficult problem of sending a truck-full of hazardous waste from Ottawa, Ontario, where most of Canada's politicians reside, to Swan Hills, Alberta, where Canada's only integrated special waste management facility is located (or vice versa, depending on what you wish to achieve). What route should this truck take? The answer will depend on what we consider as relevant objectives in this decision problem.

Cost is clearly relevant. Hence, we may wish to find the least-cost route. The least-cost route may be the same as the shortest route, if the highway network under consideration is relatively uniform in terms of surface quality and speed limits. However, if there are different classes of highways with different speed limits, the shortest path may go through long stretches of secondary highways and take up more time than some other path that takes primary highways.

Clearly, travel cost is related to time, as well as distance. Hence, we may have to consider a combination of these attributes to find the least-cost route. If we have access to a detailed highway map, we can accomplish this task by doing some elementary calculations.

What else is relevant in this decision? Clearly, the shipment of hazardous waste will carry some risks to the public along the route, as well as to the environment. Accidents do happen, and the contents of the truck can be released if the accident is serious enough. An accident can be quite costly to your company in several ways; for example, in addition to the costs of delays, you may be liable for the cleanup costs and damages, and your insurance premiums are likely to increase. Hence, we would like to consider minimizing risk, along with cost.

How should we quantify risk on a stretch of highway? Do we focus on the likelihood of an accident by the truck, or on the consequences of that accident, or both? Should we use historical accident rates on that highway (or highway class)? What about known danger spots, such as railroad crossings and left turns? To evaluate the population to be impacted, should we use the number of people that live within a certain distance of the highway? If so, what should be this distance? How do we include buildings with a high concentration of population (such as schools and hospitals) in our analysis? How do we consider environmental risk? It seems that the problem of selecting the "best" route for our hazardous waste shipment between Ottawa and Swan Hills is more complicated than the selection of the route for our long weekend trip.

Sources of complexity
There are two major sources of complexity in this problem:

quantifying different objectives, and
trading off the objectives.

The first one is more of a technical problem; a problem of identifying the necessary data, collecting and processing it, and identifying certain good routes via the use of quantitative modeling techniques. The second one is more of a judgment problem. Since the problem has several objectives, it is quite likely that that there will not be one "best" route. A short route may trace highly populated areas of the country; a route that goes through sparsely populated areas may involve stretches on narrow undivided highways with high accident rates. Hence, some subjective value judgments have to be made, which depend on the prevailing values of the company and society.

It is possible to eliminate the first source of com plexity to a great extent, and the second source of complexity to some extent, by the use of a computerized decision-support system. Furthermore, without such a system, it is very difficult to make an intelligent and defensible decision regard ing the shipment of hazard ous waste which considers all relevant factors.

In case you do not like hypothetical examples, let us switch to an example from recent Alberta history to make our point. Originally, the Swan Hills special waste treatment facility's mandate was to treat only those hazardous wastes generated within Alberta. In 1994, the Alberta government considered a proposal from the operators of the Swan Hills facility whereby hazardous waste from eastern Canada would be shipped to Swan Hills for treatment there.

The Alberta government funded a risk assessment study to evaluate the additional risk imposed by this "foreign" hazardous waste traffic destined for Swan Hills. This study, conducted by the Institute for Risk Research of the University of Waterloo, found that the additional risk was negligible. This was a fine piece of research and considerable effort had been spent on collecting reliable data on highway accidents. However, we noticed that this study focused on accident probabilities and viewed an increased number of traffic accidents as the only undesirable consequence of these hazardous waste imports.

The substance under consideration at the time (PCBs) does not pose an immediate danger to humans in case of a spill. However, a truck accident may result in an open fire, and PCBs produce dioxins and furans in an open (i.e., low temperature) fire. Although the scientific jury on dioxins and furans is still out, there is considerable evidence that they are carcinogenic. Transport Canada regulations require that the immediate vicinity of any PCB accident resulting in a fire be evacuated. Hence, a PCB fire in a large city may result in the evacuation of a large number of people, and we believed that this should have been included in the analysis.

Decision-support tool
In a separate study, we decided to quantify the risk of evacuation due to a PCB truck fire. For this exercise, we needed to obtain detailed population data (aggregate data for small towns, district-based data for larger cities). We also had to develop some methodology to go from population density data to a count of people to be evacuated in the case of a PCB fire at some point on the highway. This was a tedious, time-consuming task. By the time we were ready to publish our results, the political decision to allow imports of hazardous waste had already been made.

If a computerized decision-support tool had been available to us, we could have completed the missing part of the study very quickly, and provided additional information to those players involved in making the decision. Likewise, any interested individual could have used the software themselves to carry out their own analyses. We believe that such a decision-support system, which allows for intelligent participation in the political process by all interested parties, has the potential for facilitating negotiations between groups with different agendas, and for facilitating decision-making by consensus. This does not make a case against hazardous-waste imports into Alberta, but a case for informed decision-making.

It is worth mentioning that the Waterloo study focused on the Alberta segment of the PCB imports only. You can imagine that this is only a small part of the trip. However, it seems to me that an evaluation of the risks of transporting hazardous waste all the way from Ontario and Quebec to Swan Hills is useful in national and provincial debates about hazardous waste. Unfortunately, such studies require considerable effort since there exists no computerized decision-support tool to facilitate such analysis in Canada.

I can list other possible uses for such a tool, such as designating dangerous goods routes in provinces, assessing the environmental risks -- particularly for sensitive areas -- of dangerous goods transport, and facilitating the location of an integrated special waste management facility in an eastern province in Canada (which would, in turn, considerably reduce transport risks).

To summarize, currently there is no user-friendly decision-support system for hazardous materials transport decisions in Canada. What about the United States? Well, there are at least two commercial products south of the border. I have been fortunate enough to be provided with a test-copy of one of them (strictly for academic use) for the last two years. I will use an example from the test-copy to demonstrate the potential use of such a DSS for a carrier.

Hazardous trip
Suppose you wish to send a hazardous materials truck from Chicago to Nashville, Tenn. Which route should the truck take? Four routes are displayed in Figure 1. From left to right, the routes are:

the minimum population exposure route (assuming a one-mile impact zone on both sides of the road),
the minimum accident frequency route,
the shortest route, and
the most practical route.

The most practical route is found by considering trip length and trip duration simultaneously. It represents the route a non-hazardous truck is most likely to take.

Note that, spatially speaking, the four routes have very little in common. The shortest route (Route 3) uses divided highways, such as U.S. 41. The minimum-accident route (Route 1) and the most practical route (Route 4) select interstate highways, such as I-57 and I-65, where the accident rates are lower, and the minimum-population route (Route 1) wiggles through primary and secondary (undivided) highways, such as Illinois Route 3 and Kentucky Route 164.

The tradeoffs between the different relevant criteria are also non-trivial. For example, when we compare Routes 1 and 2, we find that Route 1 is five times as accident-prone as Route 2, while Route 2 exposes twice as many people to risk as Route 1. Route 2 is about 20 percent longer than Route 3, but is associated with 50 percent less accident probability than Route 3. Route 1 is twice as long as Route 3, but exposes only 40 percent as many people to risk as Route 3. The software in question (PC*HazRoute) generates all this information, and more, to facilitate the selection of a route between Chicago and Nashville. Of course, a decision-maker is quite likely to choose none of these "optimal" routes, and to go for some other "compromise" route instead.

While they are very useful tools for day-to-day route selection purposes, the existing commercial products in this area cannot be conveniently used for strategic decisions. For example, suppose a state (or a group of states) wants some assistance in locating a low-level nuclear waste repository. Clearly, any candidate location is associated with a set of feeder routes. It is safe to assume that the decision-makers would be interested not only in the location of the facility, but also in the feeder network that would result from a facility location. They may, for example, be concerned about the spatial dis tribution (or concentration) of transport risks. They may also be willing to construct some new roads, if risk reduction is sufficiently substantial.

While the risk assessment issue is non-trivial (and still requires some research), if we assume that we have a way of quantifying risks, the problem described above is a multicriteria location/routing problem, possibly with a network design element. It has an in teresting decision-analytic component, as well as a combinatorial component. Needless to say, no existing software can handle this problem easily. In fact, this generic problem is open for research.

GIS potential
The potential for geographic information systems (GIS) in future decision-support systems in this area is very clear. A GIS would allow for the easy extraction of necessary data for various purposes. For example, we could compute the population within one kilometer of all links of a transport network with a click of the mouse. This turns the population exposure minimization problem into a shortest path problem. (Some of the existing GISs already have this feature.)

Likewise, we could classify the links of the transport network according to whether they pass near an aquatic ecosystem or not, with a simple command. This turns the problem of finding the route that minimizes the probability of a spill near an aquatic ecosystem into another shortest-path problem. To give a different example, a GIS would facilitate the implementation of a physic`l dispersion model to accurately estimate the population to be impacted in the case of a gaseous leak for given weather conditions. Such data processing power would allow us to generate input data for a variety of interesting optimization problems.

To capitalize on the potential of GIS, a DSS needs a library of optimization routines to solve problems of interest in hazardous waste logistics. Possible modules are risk assessment routines, shortest path routines, facility location routines, location/routing routines and mode selection routines. Of course, these modules must interact. The GIS would be very useful in displaying the results of the optimization modules (i.e., routes, locations, transport modes) graphically. Finally, the ideal DSS would have a multiobjective decision-support module which would facilitate the trading-off of conflicting objectives in these decisions.

Having outlined a "pie-in-the-sky" system (an OR-based DSS embedded in a GIS "shell," served with an MCDM "topping"), I would like to turn to more modest steps and summarize some of our recent research in this area. I can refer those interested in applications of operations research in this area to a recent survey paper [1], which also contains suggestions for future research.

In another recently published paper [2], we outlined a risk assessment method to accurately estimate the population to be impacted in the case of a hazardous materials accident when traveling through a large city. When an accident occurs on a highway near a small town, the entire town may have to be evacuated. This simplifies the risk assessment. However, most hazardous materials accidents in a large city would only require partial evacuation, which requires the use of zonal population distributions. Furthermore, residents living closer to highways are more likely to be evacuated than those living further away. These features complicate the risk assessment somewhat.

Modeling transport risk
We have surveyed the different ways of modeling transport risk for dangerous goods [3], and have studied the differences and similarities between the routes found when using different risk criteria, such as population exposure minimization and accident probability minimization. We found that, as in the example above, the similarity between routes minimizing different criteria is very low, and the optimal route, according to one criterion, has very low tolerance for other criteria. The level of disagreement between the different criteria underscores the need for careful quantitative analysis, particularly in the selection of the proper objective.

We have developed a model to optimize route selection with consideration for insurance costs [4]. We have argued that operators should consider expected increases in insurance premiums when making routing decisions, and have showed that the inclusion of such costs may have an influence on the route selected.

In another recent paper we have evaluated the cost-risk tradeoffs for rail transport of dangerous goods in the United States [5]. It seems that significant reductions in risk can be achieved in return for modest increases in costs in certain cases. On average (based on 24 randomly-selected origin-destination pairs), a 1.5 percent increase in route length results in a 12 percent reduction in societal risk (which is defined as the product of incident probability and exposed population), and a 5 percent increase in length results in a 23 percent reduction in risk.

We have also explored a population-constrained, shortest-path model in the context of high-level nuclear waste shipments [6]. As one would expect, this model avoids large population centers. It is possible to reduce the maximum population exposure for a given origin-destination pair by an order of magnitude at a negligible increase in the tour length. Avoidance of large population centers is a strategy that is likely to be followed in the final selection of these routes. Using a DSS automates the procedure and allows one to quickly evaluate different scenarios.

Finally, we have focused on the undesirability of low probability-high consequence events and have suggested several catastrophe-avoidance models [7]. One of these models deals with the variance of the impacts (such as evacuations), as opposed to the more popular measure of the expectation of the impacts. We have suggested a simple way to find the minimum-variance route, and provided a numerical experiment to emphasize the difference between this criterion and other more traditional ones. This is an example of basic OR research that could be incorporated into a future DSS for hazardous materials transport.

My goal is to build a prototype of a DSS for hazardous materials transport (my "pie-in-the-sky" system) within the next three years. Clearly this is a multidisciplinary project that requires a team approach. Currently, we have compiled a Canadian-American team made up of seven core individuals (John Hodgson, Gilbert Laporte, Michel Gendreau, Teodor Crainic, Rajan Batta, Ted Glickman and myself) and a significant amount of funding from the Natural Sciences and Engineering Research Council of Canada with which to "bake" our pie. Now we just have to come up with the right recipe.

References

E. Erkut and V. Verter (1995), "Hazardous Materials Logistics," in "Facility Location: A Survey of Applications and Methods," a Springer-Verlag book edited by Zvi Drezner.
E. Erkut and V. Verter (1995), "A Framework for Hazardous Materials Transport Risk Assessment," Risk Analysis, Vol. 15, No. 5, pp. 589-601.
E. Erkut and V. Verter (1995), "Modeling of Transport Risk for Hazardous Materials," Research Report 95-2, Dept. of Fin. and Mgmt. Sci., U. of Alberta.
E. Erkut and V. Verter (1995), "Hazardous Materials Routing under Insurance Costs," Research Report 95-3, Dept. of Fin. and Mgmt. Sci., U. of Alberta.
T. Glickman and E. Erkut (1996), "The Tradeoffs Associated with Risk-Conscious Routing of Trains with Hazardous Freight," Research Report 96-2, Dept. of Fin. and Mgmt. Sci., U. of Alberta.
E. Erkut and T. Glickman (1996), "Minmax Population Exposure in Routing Highway Shipments of Hazardous Materials," Research Report 96-1, Dept. of Fin. and Mgmt. Sci., U. of Alberta.
E. Erkut and A. Ingolfsson, "Catastrophe Minimization in the Routing of Dangerous Goods," presented at the INFORMS meeting (November 1996).

Erhan Erkut is the Alexander Hamilton Professor of Management Science in the Faculty of Business, University of Alberta.

For more information, put the number 2 in the appropriate space on the
Reader Service Form

E-mail to the Editorial Department of OR/MS Today: orms@lionhrtpub.com

Lionheart Publishing, Inc.
2555 Cumberland Parkway, Suite 299, Atlanta, GA 30339 USA
Phone: 770-431-0867 | Fax: 770-432-6969
E-mail: lpi@lionhrtpub.com