Louise K. Comfort and Shi-Kuo Chang
The proposed research addresses four criteria of high performance computing and communications. First, the research addresses the national need to increase the efficiency of preparedness, response, and recovery operations in disaster management. Disaster management represents a set of interdependent problems that require intensive communication and coordination among organizations and jurisdictions to reduce risk and losses to the nation. Unresolved, these problems become cumulative and losses increase exponentially. Recent disasters have demonstrated that investment in information technology significantly increases the capacity of local communities to respond in timely, effective ways, but the technical information infrastructure has not been developed to extend this capacity to the national level. This research proposes to investigate the design and development of a distributed, intelligent, spatial information system (DISIS) that will assist practicing managers in solving the problems of disaster management at local, state, and national levels of jurisdiction.
Second, the proposed research specifically addresses the problem of hazardous materials management, but has implications for managing the entire range of hazards that afflict the nation: earthquakes, floods, fire, and severe winds. Managing the toxic products of our industrial society affects every US community on a daily basis. The steady accumulation of hazardous materials from industrial production and practice represents a continuing and growing threat to the environment, health, and safety of the nation. Current laws governing hazardous materials management are based upon a "community's right to know" of the presence of these materials, but the complexity of tracking the quantity, type, direction, and timing of 466 extremely hazardous materials flowing among some 86,000 communities, each with varying numbers of facilities for use and storage, throughout the nation rapidly exceeds the capacity of conventional means of monitoring. Further, the problem is dynamic, as monitoring involves both storage at fixed facilities and transportation of hazardous materials via road, rail, water, and air. These information requirements can only be met using by a carefully designed high performance computing and communications infrastructure.
Third, the number of communications within and between organizations and jurisdictions increases enormously in response to a major hazardous materials release or natural disaster, overwhelming the capacity of conventional technologies to receive, process, and transmit information in a timely, accurate manner. DISIS would provide a means of facilitating the transition between routine and disaster operations, enabling communities to assume greater responsibility for their own safety.
Finally, the University of Pittsburgh has advanced facilities in computing and experienced faculty in computer science, public management and policy, and engineering. DISIS represents a major cooperative effort of public, private, and nonprofit organizations to address shared risk. Once achieved for hazardous materials, gains made from this effort may be extended to the dynamic and uncertain conditions of natural hazards.
This research proposes to investigate the design and development of a distributed, intelligent, spatial information system (DISIS) that will assist practicing managers in the problem solving environment of disaster operations. The prototype addresses the problem of decisionmaking under dynamic conditions in which systems of organizations -- public, private, and nonprofit -- evolve rapidly in response to a major natural or technological disaster and then contract as needs are met and participating organizations return to their normal routines. This sequence of rapid transitions requires timely, accurate search, processing, and integration of information from multiple sources and multiple disciplines to enable the practicing managers to mobilize response operations quickly, to allocate scarce resources of equipment, personnel, and supplies efficiently, and to de-mobilize them responsibly as needs change.
The demands for rapid processing of information from multiple sources under the urgent constraints of a natural or technological disaster create a stressful, dynamic operating environment for practicing disaster managers, who bear legal responsibility for protection of life and property in their respective communities. Advanced computing techniques of information search, processing, representation, transmission, and logical inference from known data to possible courses of action provide the potential for significantly increasing the capacity of multiple organizations in a community to reduce risk from known hazards and to coordinate response actions more effectively when disaster does occur. Carefully designed, a network of computers operating in parallel can support a network of organizations engaging in parallel response operations activated by the same hazardous event.
The technical computing and information infrastructure enables the set of practicing organizations to mobilize coordinated actions quickly and efficiently in response to a given hazard within a single jurisdiction, between jurisdictions as the response expands, and within and between jurisdictions as the response de-escalates and turns to recovery. It also enables organizations to learn more effectively from the shared experience.
The DISIS prototype is designed to support decision-making by practicing managers through all phases of disaster operations -- mitigation, preparedness, response, and recovery. While the DISIS may be developed for all hazards -- earthquakes, fire, flood, or severe winds -- this prototype focuses on hazardous materials management as response to a release moves through organizational and jurisdictional levels. Hazardous materials management involves the reduction of risk and response to releases deriving from the production, storage, and transportation of hazardous materials, a risk borne by all US communities. Project Year I will focus on interactions between organizations with distributed information at the city/county level, PY II on information processes among counties at the state level, and PY III on interactions escalating and de-escalating among local, state, and national levels of jurisdiction. These processes include monitoring changing states of risk at multiple locations for disaster managers with varied levels of training and responsibility and providing timely feedback on a national scale.
The DISIS prototype is designed to support a sociotechnical dual use system for disaster management. That is, managers may use the system in their respective public, private, and nonprofit organizations for daily, routine operations in managing hazardous materials: reporting what quantities of which products are used, produced, stored, or shipped within their respective areas of responsibility. When a release occurs and threatens other segments of the community, the DISIS activates a community-wide response network that creates a disaster- specific knowledge base by drawing relevant information from multiple data bases and multiple organizations to support interorganizational and interjurisdictional decision making in reference to that event.
The prototype links two types of information processing systems: 1) the technical system of computers; and 2) the human cognitive system of decision makers and their respective organizations. Both systems are amplified by networks of communication among multiple computers for the technical system and multiple decision makers for the organizational system. The load, rate, and complexity of information that is transmitted within and among organizations in dynamic disaster operations is massive. Without technical support, the information processing demands overwhelm the cognitive capacity of individual managers and organizations to absorb, process, and use the flood of incoming information as a basis for timely, informed action. Using distributed knowledge bases and a network of computers operating in parallel, the technical computing system is designed to support the human organizational system in its conduct of distributed, parallel activities in response to the hazardous event.
DISIS provides information search, processing, representation, storage and retrieval functions with electronic communication, graphic mapping, and logical inference capabilities. The functions are:
1. An interactive, electronic field status board
2. A graphic mapping capability
3. A computerized capacity for logical inference
These functions can improve the utility of information available to emergency managers engaged in separate but related tasks vital to emergency operations. The field status board uses the concept of an electronic blackboard to enable emergency managers to report changing conditions from multiple field sites to an emergency coordinating center. This information is integrated by computers in a continuously evolving, emergency-specific database of events, conditions, actions, outcomes, resources, and problems that can be accessed directly by authorized emergency managers from remote sites.
Using a graphic mapping capability, information from the field status board can be displayed graphically at remote sites, enabling managers at distant locations to visualize operating conditions in the emergency environment. Using the active indexing technique, this information can be updated on the maps as conditions change.
Data from multiple sources can be used with computerized logical inference routines to produce a calculated set of alternatives for response under specified conditions. Emergency managers can use such routines to explore alternative actions or to confirm possible choices against existing data from the knowledge base. These three functions produce information that is stored in a layered, multijurisdictional knowledge base by type of task, discipline, and time phase in disaster operations.
The prototype DISIS allows emergency managers to order, store, recall, and exchange information relevant to hazardous materials management in three different ways: 1) within organizations; 2) across organizations within jurisdictions; and 3) within a network of organizations that crosses jurisdictions. This capacity enables emergency managers working in positions of varied responsibility within the emergency management system to build a timely, shared information base in reference to a specific threat or release of hazardous materials.
Two innovative techniques will be incorporated into the DISIS prototype. First, active indexing represents an important innovation that allows practicing managers to update their operational maps with incoming information as conditions change in the dynamic disaster environment. The active index facilitates the accessing and automatic manipulation of visual objects in a distributed environment. With an active index, we can effectively and efficiently handle visual objects that respond to accessing, probing and other actions. In conventional database systems, keyword-based indexing techniques are adequate to support users' needs. In visual information systems, there are many applications that cannot be properly supported by keyword-based techniques. Users often want to access/manipulate visual objects by shape, texture, spatial relationships, etc. (Chang and Hsu, 1992).
The active index is a dynamic index structure that is activated when a message is sent to that type of index cell. Once activated, an index cell will remain active, until either its useful life time expires, or it has entered a "dead" state. Index cells can be activated at different nodes in a distributed system to respond to the requests (messages) from the user. The active index structure facilitates distributed knowledge management and allows the parallel computation and retrieval of information.
The basic concept of the active index is to respond to the environmental changes and take corresponding actions according to user defined knowledge. An active index is a set of index cells (ICs) connected by messages (Chang, 1995). An IC accepts input messages and performs some computation. It then activates another group of ICs, and posts the output message to these output ICs. If some of the output ICs have already been activated, they may simply accept the output from the current IC. The first output cell that accepts the output message will remove it from the output list of the current cell. After its computation, the IC may remain active (live), or de-activate itself (dead). An IC will become dead, if it remains inactive for a certain period of time, i.e., if no other cells (including itself) send messages to it. An active index consists of a finite number of ICs. When the active index is in actual computation, it consists of a time-varying collection of ICs in different states, accepting certain input messages and posting output messages to the output lists.
Each IC has two functions: acceptance and knowledge. The acceptance function, denoted as f, determines when the IC is enabled and ready to fire. Once all the required messages become available, f will remove the messages from the output lists of the message-sending ICs and enable the IC. The knowledge function, denoted as g, performs the firing procedure for the IC. Once the f enables the IC, g will take over the control and fire the IC. According to the messages accepted by f, the firing procedure will decide: (1) the next state of the IC; (2) how to generate new messages for specific ICs; and most importantly (3) how to perform a specific action sequence.
The active index contributes two important features to DISIS knowledge management: `active' and 'private'. The users can specify their private knowledge and then combine that with the system's knowledge, resulting in greater flexibility in DISIS's adaptive behavior. The `private' knowledge also means polymorphism - certain objects can obtain reactions different from reactions to other objects even in the same environment. For emergency message management, we can, for example, order messages by the importance of message classes so that if the recipient fails to view a particular message, a reminder will show up on both the sender's and the recipient's screens. Another interesting example is the inclusion of different levels of pre-fetching methods in DISIS. DISIS will provide the basic level of pre-fetching method. For a particular application, users can add their own pre-fetching methods based on their special considerations.
In order to combine private knowledge with the system's knowledge, we can divide the ICs in the active index into groups to form a hierarchy. In this hierarchy, one class of ICs can share the same methods. Messages sent to higher level ICs will be handled by the higher level methods. Only when there is no higher level method will the message be sent to ICs at lower levels and handled by lower level methods (Chang and Hsu, 1992; Chang, Hou, and Hsu, 1992; Chang, 1995).
Second, logical inference using fuzzy language and probability estimates will be used to capture the anticipatory logic characteristic of practicing managers who seek to bring a dynamic, uncertain set of conditions under control. This prototype will integrate the technology of artificial intelligence with conventional databases to achieve a more powerful information system for decision support processes and training in hazardous materials planning and response. Since the information items in a disaster situation are often imprecise and incomplete, it is necessary to use fuzzy linguistic variables to describe the situation. The information stored in the database can also be imprecise and incomplete, requiring the use of fuzzy algorithms to support decision making. For example, the fuzzy algorithm computes the significance level of an environmental threat. If the significance level exceeds a predefined threshold, this is regarded as an event, leading to the activation of the active index to perform certain actions.
We envision the following approach. A blackboard system is used so that every piece of information is first posted on the blackboard. Fuzzy inference algorithms (as well as other inference algorithms) are applied to generate de- rived information which is again posted on the blackboard. The information item on the blackboard may activate index cells which again activate other index cells, thus creating an active index. Since the blackboard system exists at each node in DISIS, there are index cells at each node, creating an index net spanning the entire distributed system.
We illustrate the use of an active index in reference to a specific disaster event, the Pittsburgh Oil Spill, in terms of the participating actors, frequency of interaction among the actors, goal of the operations, and principal opera- ting conditions. The following scenario, drawn from actual events, will be used to illustrate the applicability of the active indexing technique to an evolving set of disaster operations (Comfort, Abrams, Camillus, and Ricci, 1989).
On January 2, 1988 at 5:10 p.m., a four-million gallon tank of diesel fuel collapsed at the Ashland Oil Company's tank storage site on the Monongahela River 27 miles south of Pittsburgh. Approximately 3.8 million gallons of diesel fuel No. 2 were in the tank, and force of the collapse caused the fuel to splash out of its containment area into the containment area of the neighboring Duquesne Light plant. In sub-freezing temperatures, diesel fuel No. 2, with a flash point of 50 degrees, presented little danger of fire. With cautious relief, emergency operations personnel began to organize the massive clean-up operations, expecting to contain the spill at the Ashland site and adjacent properties.
At approximately 10:00 p.m., emergency response personnel, making a routine check of the spill site, discovered gasoline leaking from a nearby tank. The first tank's collapse had damaged a second tank, filled with gasoline, and caused at least four leaks in its piping structure. Leaking gasoline, with a much lower flashpoint than diesel fuel No. 2, created a more urgent danger. Emergency response personnel focused their attention and resources on identifying and plugging the leaks in the gasoline lines, fearing an explosion that would threaten the 700 residents of the Town of Floreffe, just across the highway from the Ashland Oil Company facility. Virtually all work on clean- up operations stopped as local officials ordered the evacuation of 1200 residents of Floreffe and adjacent areas as a cautionary measure. Working through the night, emergency response personnel found and plugged the last gasoline leak at dawn, and evacuated residents were allowed to return to their homes, weary, but out of danger.
At first light, emergency personnel discovered a third, and potentially more serious, danger. Throughout the night, the spilled diesel fuel had flowed into the Monangahela River through an undiscovered storm drain located on the Duquesne Light property next door. The spill, which emergency personnel had expected to contain on land, had now created an oil slick on the Monongahela River that extended bank to bank, seventeen miles long. The Monongahela River serves as the main source of water supply for some 850,000 residents in the Pittsburgh Metropolitan Region. Ordinarily, diesel fuel would float on top of the water, not endangering the water intakes located some 17 feet below the surface of the river. The fast-running Mon, however, had carried the slick over two locks and dams, and the tumbling action of the river had emulsified the oil through the water to the depth of the water intakes.
Emergency personnel confronted the threat of contaminating the water supply of 850,000 area residents or shutting down the water intakes, limiting severely the supply of water available to residences, businesses, hospitals, schools, and other facilities in the area. Given the added risk of permanently damaging the water filtration systems on the river, the water authorities closed the water intakes, cutting off the water supply to several large municipalities in the area. Lack of water created a new threat to public safety, as the fire departments in the region were dependent upon water for fire suppression.
This set of actual events illustrates the type of dynamic, evolving crisis characteristic of interdependent systems in which active indexing would enable emergency managers at different locations to visualize the different conditions contributing to the escalating event in a more accurate and timely way.
The design of the DISIS prototype requires advanced skills and knowledge in both organizational and technical disciplines. Disaster environments are governed by legal requirements and professional standards of disaster management that represent a specialized field within public policy, management, and organizational design. The technical requirements for building the DISIS prototype, in turn, involve advanced knowledge and skills in computer science. Building the knowledge base regarding the actual infrastructure within which hazardous materials are stored and transported requires specialized knowledge and skills in engineering. Each of these disciplines are critical to the informed, effective development of this prototype, and each is represented among the project investigators and graduate student research staff.
The proposed research is designed to implement the prototype in a trial demonstration in hazardous materials management over a three-year period. We are fortunate to have the full cooperation of the Allegheny County Emergency Management Agency, located in Pittsburgh, Pennsylvania, for the initial development of the DISIS prototype at the local jurisdictional level. The project also has the active support of the Environmental Systems Research Institute in our implementation of the active indexing technique using the ArcInfo software. We will have the expert advice of consultants at Lawrence Livermore National Laboratory, Livermore, California to guide us in the design and implementation of a prototype for national implementation.
The trial demonstration would allow us to measure the rate of increase in communication and coordination among organizations made possible by their use of the DISIS prototype and to measure whether increased communication and coordination produced the expected improvement in efficiency and organizational performance in managing hazardous materials.
The proposed research will allow us to build a sociotechnical system that can shift quickly from routine tasks of management by single organizations to create networks of organizations and jurisdictions operating as a coherent, coordinated disaster response system during the immediate, urgent need following a toxic release or hazardous event, and returning to routine tasks when the threat is brought under control. These findings will support management and training in emergency response and recovery organizations as well as contribute to our theoretical understanding of processes of transition and self organization in both technical and organizational systems.
Chang, S. K. and A. Hsu. 1992. "Image Information Systems: Where Do We Go From Here?" IEEE Transactions on Knowledge and Data Engineering. Special Issue Celebrating the 40th Anniversary of the Computer Society. (October):431-442.
Chang, S. K., T. Y. Hou, and A. Hsu. 1992. "Smart Image Design for Large Image Databases", Journal of Visual Languages and Computing, Vol. 3, No. 4, (December):323-342.
Chang, S. K. 1995. "Toward A Theory of Active Index", Journal of Visual Languages and Computing, Vol. 6, No. 1 (March):99-116.
Comfort, L. 1994. "Self Organization in Complex Systems." Journal of Public Administration Research and Theory, Vol. 4, No. 3 (July):393-410.
Comfort, L., J. Abrams, J. Camillus and E. Ricci. 1989. "From Crisis to Community: The Pittsburgh Oil Spill" in Industrial Crisis Quarterly, Vol. 3, No. 1, : 17-39.
Comfort, L., T. Woods and J. Nesbitt, "Designing an Emergency Information System: The Pittsburgh Experience" in Tom Housel, ed., Advances in Telecommunications Management, Vol. 3 (Greenwich, CT: JAI Press, 1990): 13-31.