Simulation based emergency response planning
Toros Caglar, PhD Student
School of Business, The George Washington University, USA
The awareness of the need to be prepared for emergencies has increased in the recent years with occurrences of incidents such as 9-11 and Hurricane Katrina. Emergency preparedness requires developing well thought out plans for use should the need for emergency response arise. The plans can be done at high level only given the number of factors that may be different for a particular emergency. The high-level plans may be used as a guide, but a unique plan will need to be rapidly developed for a coordinated response to an actual emergency. Such emergency response planning is highly demanding. Any sizable emergency incident would require an almost immediate plan to be provided by the emergency dispatchers, followed by an updated plan that may be generated by multiple people in the relevant emergency operations center (EOC). In both cases, the planners have to comprehend a range of factors including:
- Information on the emergency incident and the current situation,
- Expected changes in the situation,
- Availability of resources across a multitude of agencies,
- Any guidance available from applicable plans,
- The applicable chain of authority, and
- A severe time constraint.
The planners have to rapidly come up with a plan in consideration of the aforementioned factors and communicate it to all the agencies involved in the response. The situation being highly dynamic, frequent updates may be required to the emergency response plans.
Emergency response planning can be considered as a project planning endeavor since it is a temporary endeavor undertaken to create a unique set of services (Jain, 2006). There are differences in terminology including referring to the project plan as an incident action plan or an emergency response plan, referring to the project manager as the EOC manager, etc. The response planning is usually done by highly experienced emergency response professionals. Even with long experience of participating in emergency response, the task is a huge challenge given the time constraint and the other factors that need to be taken into account as mentioned earlier.
We propose a simulation-based approach to emergency response planning to assist the personnel involved in such roles. Our proposed approach relies on quick access to the most recent data available on the situation including the status at the incident site and of available resources. It utilizes predefined policies and procedures to rapidly develop an emergency response plan. The approach can help save precious time in an emergency response situation. The EOC manager and other personnel can review the system generated plan quickly rather than spending time on coming up with one from scratch.
In the next section we review the literature relevant to emergency response project management. We present our proposed approach in detail in section 3 followed by an application example based on a hypothetical scenario in Washington DC. We discuss the pros and cons of the approach based on the learning from the application example in section 4. We conclude the paper in section 5 with directions for future research.
Most of the research in the area of emergency response has concentrated on higher level planning issues such as facility location and resource allocation (Simpson 2006). Resource allocation in emergency response has been studied for decades (see for example, Ignall, Carter & Rider, 1982). We provide a few recent examples of such efforts. Gong and Batta (2007) present an iterative procedure for allocation and reallocation of ambulances to casualty clusters in a disaster relief operation. Gupta and Ranganathan (2007) present a non-cooperative multi-step games approach to allocate resources among multiple crises in an urban environment.
Few efforts have defined emergency response as a project. Jain (2006) identifies emergency response as a project and proposes a parametric approach to plan such projects. Here we suggest an alternative simulation-based approach to the development of a project plan for emergency response. Simpson (2006) also identifies emergency response as a project and goes on further to classify them as a specific class of projects defined as hyper-projects. Hyper-projects are defined as possessing three additional characteristics in addition to those in a regular project: an external pacing function, network scalability and pre-emptive tasks, and expediency. He presents a symbolic model for modeling the initial part of a residential firefighting effort and shows that its results compare favorably with a spreadsheet model. He recommends using modeling and simulation for the study of scalable network portion of the hyper-project network. Our proposed approach suggests a way of using modeling and simulation for planning the emergency response project through a complete hyper-project network, including the initial and the scalable portions.
Simulation has been used as a tool for evaluation of plans and policies for a long time. Carter and Ignall (1970) compare different policies for locating and dispatching fire fighting units using simulation models. Their results were used by the Fire Department of the City of New York. Mould (2001) employs discrete event simulation to estimate the effectiveness of emergency evacuation procedures for offshore oil structures in adverse weather conditions.
Simulation has also been used specifically for evaluation of project plans, in particular, for impact of stochastic factors. AbouRizk and Wales (1997) presented a combined discrete event and continuous simulation based methodology for evaluating construction project plans.
Researchers have suggested use of simulation to provide decision support to emergency planners. De Silva and Eglese (2000) developed a spatial decision support system for emergency planners based on a combination of simulation and spatial data handling and display capabilities of a geographical information system (GIS). The system was aimed for use by emergency planners for experimenting with different evacuation procedures in order to plan for various contingencies. Our approach, though independently developed, is similar to the work by De Silva and Eglese in concept and goes further to integrate the system with a project management software.
We propose the use of simulation to generate project plans for emergency response. Our approach was motivated by use of simulation for generating production schedules. Such applications of simulation developed in academic research initially and are well established in commercial tools now. The approach needs to be adjusted to the different characteristics of emergency response than a production environment. In the next section we describe our approach and its implementation.
Our approach is based on use of discrete event simulation for generation of project plans for emergency response. Banks (2000) defines discrete event simulation as “the imitation of the operation of a real-world process or system over time. Simulation involves the generation of an artificial history to draw inferences concerning the operational characteristics of the real-system that is represented.” Typically, simulation is used for evaluation of alternative policies and procedures. The operation of the system under a selected policy is imitated through simulation including modeling of stochastic factors and an artificial history generated. The artificial history is then analyzed to measure the performance of the system. Multiple simulation runs are executed using different initial settings of the stochastic factors to model the real life system's operation under a wide range of conditions. The outputs of the multiple runs are analyzed to draw statistically correct inferences on the performance of the selected policy. The set of runs can be repeated for alternative policies and their performances can be compared to identify the best performing option.
The capability of simulation to generate the artificial history of the modeled system provides the opportunity for its use for generating plans. Simulation can be set to run under deterministic settings and its output history captured to serve as the plan. In case of production scheduling, a simulation model of a manufacturing system is executed with defined policies for decision making and the resulting history is used as the schedule. The quality of the production schedule is determined by the policies used for making the decisions such as the selection of the next part to be processed from those waiting in the queue in front of a group of machines and allocations of resources, including machines and operators for each processing step for a part.
We note that the use of a sequence of events generated from a deterministic simulation for planning for real life operations requires defining ways to handle the discrepancies that would occur due to stochastic factors. For example in a production operation, stochastic events such as machine failures and yield variations will make it difficult to follow the production schedule. Ways to accommodate such changes include regeneration of the production schedule, building in time buffers in the schedule for accommodating variations, inflating the operation times of individual operations to accommodate variations, and providing rules to production floor personnel for modifying the schedule in response to such events. Given the complexity of the production scheduling problem, the generated schedule provides a good plan to serve as the basis for making modifications as needed to accommodate stochastic variations.
We propose to use a simulation model of emergency incident and involved resources to generate project plans for emergency response. Similar to the case of production scheduling, policies need to be defined to make decisions as the events unfold in the artificial environment. The decisions include determination of the number of response resources to be assigned to attend to an incident, identification of the specific resources to be dispatched, identification of routes taken by the resources to travel to the incident site, and their following courses of action. The actions of resources such as police and fire department personnel may involve staying at the incident site to execute their respective responsibilities, while for the emergency medical technicians (EMT) it may involve triage of the casualties, stabilizing actions, and transportation to the nearest hospital's emergency rooms. The actions of EMTs include further decisions such as selecting the hospital that can provide fastest attention to the victims of an emergency incident. If the nearest hospital is already crowded with other casualties, it may be more prudent to take the victims to hospitals further away. The defined policies can be applied during the simulation and the artificial history generated can be recorded for use as the emergency response plan.
Emergency response is arguably much more prone to stochastic variations than a production floor. Also, the nature of the domain does not allow the luxury of building in time buffers in the plan or to inflate the times for individual response activities to accommodate variation of real life events from those in the plan. We propose that any major unexpected events be accommodated by regeneration of the emergency response plan based on a new simulation run executed with most current information as its initial condition. Minor variations from the plan may be accommodated without making any changes.
Concept of Operations
In this section, we describe the concept of operations for a system based on the simulation-based approach described above. A system for generation of emergency project plans will be primarily useful in emergency operations centers that are responsible for coordinating the response to an incident. It may be noted that emergency operation centers are set up usually for larger than routine emergencies. The system can provide a project plan appropriate for response based on initial information regarding the incident. As the situation unfolds and more information becomes available, the system can generate new plans appropriate to the new circumstances.
The above concept assumes the availability of the following:
- A good information system infrastructure allowing close to real time data feeds from incident site in to the systems at the EOC;
- Data that describes the initial incident including an estimated severity level on a one to five scale with 1 being the most severe, an estimated number of casualties, and a breakdown of the casualties in five ESI classes;
- Data on availability of resources including those available immediately after a 911 call is received at the dispatching, and those that will become available over the time the response activities are expected to last;
- GIS representation of the area surrounding the incident including information on locations of response resources, including police and fire stations and hospitals; and
- Data on resource deployment practices followed by the emergency managers defined as the number of different crews deployed for incidents of severity ranging on a one to five scale.
On the occurrence of a major incident, initial response resources can be deployed based on the information provided by the person calling in about the incident and applicable policies. For example, reports of an explosion with multiple casualties may involve deployment of a large contingent of emergency response resources. The dispatcher can use the description provided by the initial call to determine the severity of the incident. Based on the estimate of the severity, the dispatcher can dispatch the corresponding number of response resources.
For large incidents, the information from 911 calls and other reports will trigger the setting up of an EOC. With the proposed concept of operations, the EOC displays will include a decision support system that generates the incident action plan based on simulation. The initial plan display available to the EOC personnel will be based on the initial dispatches made by the dispatcher. Once the EOC personnel are in place, they can request a draft plan based on any updates received from the scene by the first arriving responders. The draft plan can serve as a basis for quick discussions among the EOC personnel leading to a final plan for deployment.
As the situation unfolds and the response actions are implemented, variations may occur due to circumstances not comprehended in the plan. Minor variations may not require revision of the plan, but large variations would. The simulation-based decision support system can receive the inputs based on the current situation and generate revised draft plans that can be reviewed, modified as necessary by the EOC personnel, and released as the revised plan for deployment. It is clear that a strict version control procedure will need to be in place so that everyone on the team is following the same version of the plan. All responders would need to have access to mechanisms to receive updated plans quickly. These may include things such as personal digital handheld devices for personal instructions, and plug-in devices at command posts for more detailed information.
Our proposed concept of operations for the simulation-based approach for generation of emergency response project plans is summarized in Figure 1.
Emergency Response Project Plan Generation
We describe the generation of the emergency response plan using simulation in more detail in this section. The emergency response plan is akin to the project plan including a schedule of tasks and resources for accomplishing the defined objectives.
The inputs to the simulation based system include the local street map with geospatial data on emergency response locations, incident information, emergency response resource availability, and emergency response plans with added quantitative information. The local street map includes the location of responder resources including police and fire stations, current locations for police cars on patrol, and ambulances.
The incident information includes the location, the type (fire, traffic accident, plane crash, etc.), and the incident severity (a scale of 1 to 5 with 1 being the worst). Tables are provided for translating incident severity level and type of incident to number of casualties in each of the five levels of the emergency severity index (ESI) based on historical data. The ESI scale has been defined for emergency department triage by the Agency for Healthcare Research and Quality (AHRQ, 2007) with 1 representing the most urgent and 5 representing the least urgent for receiving medical care. While the initial incident data is based on historical data tied to incident severity and type, the data is subsequently updated based on reports from the incident site and used for the next emergency plan generation run.
The availability of resources will need to be defined incorporating the time window element. The minimum number of resources that are available for dispatching based on the time of the day need to be defined first. The on-call responders need to be defined next again based on the time of the day and including their expected response time. The next level will include the responders that are not defined to be on-call but can be reached and called in for larger incidents. Finally, the responders that can be called in from surrounding jurisdictions based on mutual-aid agreements should be defined with their expected response times. The availability of the resource should be defined based on time window that they can continuously work for. For a large incident, the responders may need to work through several days and the defined procedures for work and rest cycles will help define the time windows or the calendar for the resources. For example, for a wild-fire incident fire fighters may use 12 hour work and 12 hour rest cycles.
The emergency response plans are based on anticipated responses to emergency incidents of defined type and severity level. For example, a fire incident of severity level 3 (middle of the severity scale from 1 to 5) may require two fire engines and 5 police cars to be dispatched, and it may take 2 hours to extinguish the fire completely. In addition, the incident may require 5 ambulances to be dispatched for attending to the potential casualties.
The inputs are used by the simulation program to develop detailed project plans that can be used by emergency response personnel. The outputs of the simulation can be channeled through a project planning software for ease of display and modification.
The simulation system is capable of generating response plans for multiple emergency incidents that may occur close to each other. The simulation logic proceeds as follows to generate a project plan for the response effort based on the emergency response plans. The simulation initializes the location of all the resources. The occurrence of the emergency incident(s) is modeled and its location is identified. The incident triggers the attempts to dispatch the resources as required per the data based on type and incident severity level. The resources nearest to the incident location are identified and dispatched. The police and fire resources are dispatched based on the severity level of the incident. That is, the required resources are assigned to the highest severity incident first. The ambulances are sent to the incident locations based on the number of casualties needing the most urgent care. In case of multiple simultaneous incidents, the ambulances are assigned first on the basis of ESI 1 casualties across all incidents followed by ESI 2 casualties across all incidents and so on.
The simulation run models the travel of the assigned resources from their respective current locations to the incident location. The program calculates the fastest path for each resource from its current location to the incident location taking into account the congestion on different street segments. The travel of each of the assigned resources to the incident location is modeled. Once the resources arrive at the location, they stay at the location for the prescribed time to carry out their responsibilities such as extinguishing the fire. Each incident has a defined effort requirement for fire engines. If multiple fire engines are assigned to an incident, the effort is spread across them. Once the fire engines have put out the fire indicated by completion of the required effort represented by time, they indicate their availability to the dispatcher. The dispatcher may assign them to the next incident if there is a demand or send them back to their base stations. Similar logic is followed for the deployment of the police cars.
The ambulances use a different routine. Their travel is modeled similar to the other resources. However, once they get to the incident site, they model the loading of casualties into the ambulance, and travel to the nearest available hospital. Once they arrive at the hospital, they unload the casualties, and receive the destination for the next trip from the dispatcher. The ambulances repeat this cycle until all the casualties from all of the incidents have been moved to hospitals.
The execution of the actions in the simulation generates an artificial history of the events. This artificial history is sent from the simulation program to a project planning software for display as the project plan for the emergency response. The assignments used in the computer model can be provided to real-life resources for execution by the EOC personnel after any needed quick modifications.
We have developed a prototype software to implement the emergency response project plan generation capability described in the previous section. We developed the simulation model using Java and the Geotools open source library. We transfer the output of the simulation run into Microsoft Project for display and further modification.
We utilized the prototype software to develop project plans for hypothetical emergency incidents in Washington DC. We modeled three simultaneous incidents including a fire, an accident and a small plane crash in the area around the White House. We utilized publicly available data for the location of police and fire stations and hospitals and developed sample data for resource availabilities.
Figure 2 shows a screen shot of the simulation screen showing the movement of the assigned resources to the respective incidents. The screenshot includes the display of a street map of the area of interest in Washington DC. The incident locations, police stations, fire stations, and hospital locations are shown with red stars, blue squares, red squares, and blue crosses respectively. Police car, fire engines, and ambulances are represented using representative icons. The screen shot also shows the menu on the left hand side that allows capabilities such as display of different data on the map. The menu items on the top left include the capability to run the simulation. The bottom shows the message window with information on events as the simulation run progresses.
Figure 3 shows a screenshot of the generated plan displayed using Microsoft Project. The plan shows the travel of police cars and fire stations to incident locations, their scheduled time at the site, and their return back to their base locations. The plan also shows the scheduled activities for ambulances including travel to incident site, loading of casualties, travel to assigned nearest hospital, and unloading of casualties. Transferring the information into a project planning software provides additional analysis capabilities that wouldn't have been available otherwise. For example, the plan can be sorted by resource names to view the plan for each resource as shown in the figure. Decision makers can easily sort the view by time to see the simultaneous dispatches and activities going on in parallel. We can also use the project planning software to analyze the resource loads particularly after the plans have been manually modified. The plan as generated by the simulation program will not have resource overloads since the resource constraints are observed during the execution of the model, however, the manual modifications of the plan can lead to such situations.
We believe that the simulation based approach to emergency response plan generations offers multiple advantages. The prime advantage is the saving of precious time that would be otherwise required for manually putting together the information for coordinating the dispatching of resources. The ability to schedule available resources based on the data will further help in reducing the time required for decision making and deployment of resources.
We understand from popular press that there are differences in the operating cultures of the multiple agencies that are involved in emergency response. These differing cultures may lead to the difference of opinions within the EOC, even with the presence of a clearly defined command structure. Generation of the initial plan using the proposed simulation-based approach may largely reduce the discussion of alternative approaches. Of course, we realize that it may take a while for the agency representatives to agree on the emergency response plans that are input to the system. These discussions will be held in calmer surroundings and not under the pressure of an unfolding incident. As long as they might take to arrive at consensus, the emergency response plans and procedures they agree on would be very useful for future incidents. The collective input from the experts will be available for generation of initial plans in all incidents. Thus the system will allow utilizing collective knowledge of experts when an incident occurs and it will help reduce the potential for disagreement among representatives from multiple agencies.
We believe the generation of detailed project plans for emergency response will reduce the issue of which agency and which person is in charge. The current guidelines may result in transfer of EOC leadership a few times as different people arrive. As the leadership passes from one to another, they may change the plans per their perceptions. The existence of system-generated plans may reduce the potential for changes in plans based on who is in charge and even the issue itself of “who is in charge.”
We have demonstrated the feasibility of an approach using simulation to generate project plans for emergency response through the prototype and the application example. The implementation of this approach presents several challenges that need to be addressed.
The simulation runs requires a number of inputs as discussed earlier. We believe the collection of the required data and particularly the emergency response plans and procedures will require a significant effort. We also note that the plans and procedures may vary across different jurisdictions and hence we will need to develop a data-driven interface for input of such information. We also realize that in order to gain from the knowledge of experts, the system may have to capture a number of parameters and a number of emergency response plans so as to identify the applicable plan for a particular incident. We point out though that the system is intended to provide a starting version of the plan that can be updated as needed by the EOC personnel. While the system will certainly gain from larger knowledge base, it will still help save some time with generic versions of applicable detailed plans and scheduling of available resources.
We believe a more critical challenge is to gain acceptance of the approach among the emergency response community. While there have been a few major applications of decision support technology in some cities, we believe it will require a concerted effort to convince the emergency responders to place confidence in the plans generated by a simulation-based decision support system. We propose to first focus on convincing the emergency responders to test the approach in live exercises for emergency response. Such a test will allow an opportunity to demonstrate the capabilities of the system and the advantage in reacting rapidly to an imaginary emergency incident.
We have developed the system in Java partly due to the availability of open source library for geospatial purposes. We realize that, for the system to be successful, it needs to execute quickly in order to generate the initial plan and to run “what-if” analyses. We will need to make the implemented algorithms highly efficient to execute quickly, preferably in less than 1 minute for a data set comparable to the size of the application example. We may need to redevelop the simulation system in another language to achieve the desired execution performance.
We presented a simulation-based approach for generation of project plans for emergency response. We believe the implementation of the approach can help the emergency response effort through quick generation of comprehensive response plans. Simulation-based approach can be used for generating project plans for domains other than emergency response. However, a significant effort may be required to capture and collect the data, plans, and procedures to be used in the simulation model. We contend that the system will be useful even with basic knowledge inputs since it can provide a useful initial project plan that takes into account the resource availabilities. The plan can be rapidly modified by the EOC personnel to meet the needs of the specific incident.
We plan to continue to develop this approach further. We realize it is important to involve emergency response personnel and intend to do so. We developed the prototype to clearly communicate the concept to the emergency responders to evoke their interest and, over time, gain their acceptance. We also plan to address the other challenges listed earlier including collection of data and improving the execution speed and performance of the planning algorithms.
Some commercial software products are mentioned in context in the paper. This does not imply a recommendation or endorsement of the software products by the authors or their organization, nor does it imply that such software products are necessarily the best available for the purpose.
We thank Frank Riddick of the National Institute of Standards and Technology, Gaithersburg, MD, for developing the first version of the simulator used for this effort. We also acknowledge the benefit to this effort due to the availability of GeoTools open source library.
AbouRizk, S. M. & Wales, R. J. (1997, March). Combined discrete-event/continuous simulation for project planning. Journal of Construction Engineering and Management, 123(1), 11–20.
AHRQ (2007). Emergency Severity Index, Version 4. Available online via http://www.ahrq.gov/research/esi/ [last accessed on Dec. 14, 2007].
Bank, J. (Ed.) (1998). Handbook of Simulation: Principles, Methodology, Advances, Applications, and Practice. New York, NY: John Wiley and Sons.
Carter, G. & Ignall. E. (1970, October). A simulation model of fire department operations: design and preliminary results. IEEE Transactions on Systems Science and Cybernetics. 6(4), 282–293.
de Silva, F. N. & Eglese, R. W. (2000). Integrating simulation modelling and GIS: spatial decision support systems for evacuation planning. Journal of the Operational Research Society, 51(4), 423–430.
Gong, Q. & Batta, R. (2007). Allocation and reallocation of ambulances to casualty clusters in a disaster relief operation. IIE Transactions, 39, 27–39.
Gupta, U. & Ranganathan, N. (2007). Multievent crisis management using noncooperative multistep games. IEEE Transactions on Computers, 56(5), 577–589.
Ignall, E., Carter, G. & Rider, K. (1982). An algorithm for the initial dispatch of fire companies. Management Science, 28(4), 366–378.
Jain, S. 2006. A parametric approach to emergency response project management. PMI Research Conference, PMI, July 16-19, 2006, Montreal, Canada.
Mould, G.I. (2001). Assessing systems for offshore emergency evacuation. Journal of the Operational Research Society, 52(4), 401–408.
Simpson, N.C. (2006). Modeling of residential structure fire response: Exploring the hyper-project. Journal of Operations Management, 24(5), 530–541.
©2008 Project Management Institute