Dale N. Anderson, Ph.D., joined the technical staff at the Pacific Northwest National Laboratory in 1990 and is now a staff scientist and technical group lead for the Stochastic Scientific Modeling & Simulation group. His responsibilities include research and development in national security and defense programs and for the Department of Energy Ground-based Nuclear Explosion Monitoring Research & Engineering Program. He is the Pacific Northwest National Laboratory deputy laboratory lead for the program. He served for six years as the program coordinator for the mathematics and statistics program at Washington State University–Tri-Cities. He is professionally active with publications on the application of mathematical statistics to seismology and nuclear science. He completed his doctoral research in applied statistics under the direction of Barry C. Arnold at the University of California–Riverside.

Sandra E. Thompson, Ph.D., is a senior research scientist at Pacific Northwest National Laboratory. She joined the laboratory from her doctoral program in statistics at Colorado State University and has been with the lab for 3 years. She works as a technical contributor on a variety of research projects, including radiation detection, biological differentiation, and reliability issues. She was recently assigned responsibilities as principal investigator on a major national security project at the laboratory.

General Charles E. Wilhelm (retired) joined Battelle on 30 December 2002 after serving for more than 38 years as a combat arms officer in the U.S. Marine Corps and as an independent consultant on a broad range of national security, policy, and strategy issues. He is vice president and director of the Battelle Office of Homeland Security. He served as commander in chief of U.S. Southern Command and was directly responsible to the Secretary of Defense for designing and executing strategic plans addressing a full range of military contingencies and engagement activities with the 32 countries of Central and South America and the Caribbean. He is a veteran of combat operations in Vietnam, Lebanon, the Persian Gulf, and Somalia. He has been decorated by the governments of Argentina, Chile, Colombia, El Salvador, Guatemala, Honduras, Lebanon, Nicaragua, Paraguay, Peru, and Vietnam.

Ned A. Wogman, Ph.D., is director of the Homeland Security and Science & Technology Program Office at Pacific Northwest National Laboratory. He has more than 37 years of experience in national security. He is recognized internationally for expertise in the development and use of radionuclide sensors for the detection of nuclear proliferation. He served in a two-year assignment on the International Atomic Energy Agency’s Iraq Action Team to support its Wide Area Environmental Monitoring effort from 1998 to 2000; he now chairs an environmental monitoring evaluation group for the International Atomic Energy Agency. He has served as a science advisor to various Department of Energy and Department of Defense organizations and has published more than 200 peer-reviewed papers. He received his doctorate in physical chemistry from Purdue University.

Summary

Effective utilization of all available intelligence, including sensor signatures and situational awareness, is a key objective in homeland security. Binding all sources of information into an objective and lucid decision algorithm can provide clarity to identify signatures that strongly and uniquely indicate terrorist activities, thus reducing false alarms that conjure images of profiling and concerns regarding our civil rights. Our fundamental premise in this article is that the optimal integration of situational awareness, intelligence, and hard sensor signatures should begin at the field level and work backward—that is, begin with the desired outcome and work backward. Construction of in-the-field algorithms with these characteristics will necessarily be dominated by careful mathematical and scientific thought as opposed to purely empirical, unguided data analysis. The research and development effort for optimal decision algorithm construction naturally encourages homeland security communication at all operational levels, including that between scientists, intelligence analysts, government leadership, and the private sector. Why? Because decisions have consequences that impact all stakeholders, and a formal decision framework is capable of quantifying these consequences. A properly constructed framework naturally includes mathematical plug-in points for hard sensor data, intelligence, and situational awareness. These plug-in points naturally guide the formulation of information to a common standard, thus facilitating and promoting intelligence sharing. A well-established foundation to build these frameworks in the field and at the strategic level can be found in a body of theory in mathematical statistics: Bayesian decision sciences.¹ We assert that decision algorithms with these characteristics are necessary for optimal frontline operational capabilities in the war on terrorism.

Setting the Context

A police officer in a large metropolitan area responds to a call to investigate a suspicious package. The officer could conceivably have several basic pieces of information to decide on a course of action:

• an internal agency or department briefing
• hard sensor signatures (for example, chemical, biological, or radiological)
• situational awareness

In addition, the officer will have a sense of the consequences of possible actions. For example, if the package is not a threat and the officer mistakenly perceives a threat, costly operations will be set in motion.

At a port of entry into the United States, a nuclear sensor alarms a Customs officer as a tractor-trailer moves through an entry lane. The customs officer also has intelligence provided by national technical capabilities and situational awareness, possibly in the form of a license plate matched to a bill of lading. Again, each possible action on the part of the officer leads to a suite of consequences.

Each of these scenarios is quite feasible, and with well-developed protocols to guide an officer’s thinking and actions, situational awareness can in fact be used in the officer’s decisions. An officer’s thinking (guided protocols) is extremely valuable information and should not be viewed as a confounding factor in a decision. At issue is the effective, consistent, and objective utilization of this information.

The Human Sensor

Jay Davis and Don Proznitz, in an article in Physics Today,² correctly asserted that “integrating distributed sensors to reduce false positives and other extraneous signals is also a technique long used by physicists. Creating a first-rate surveillance and detection system will involve multidisciplinary technological tradeoffs as well as sociological and economic considerations.” We extend this thinking to include the human (front line) as a valuable sensor asset. Properly guided by a mathematical decision framework, the Customs officer becomes an integrated sensor, possibly requiring calibration for accuracy, but nonetheless a fully integrated sensor.

A well-established body of theory for the development of optimal in-the-field decision algorithms can be found in the methods of Bayesian statistics. The basic Bayesian algorithm involves two fundamental steps that can be iteratively repeated. In the context of homeland security, the first step is the formation of a threat likelihood or probability that is based on available information. The second step simply revises or updates this likelihood with new data—for example, hard sensor signatures. The mechanics of the algorithm can be further illustrated with an example. Each vehicle entering the continental United States through a port of entry has associated with it objective information that serves to form a likelihood of threat. Examples are car rental information, license plate number, and bill of lading. In many cases this likelihood alone could serve to triage the vehicles, but not optimally. To utilize all information, this threat likelihood is then updated with a physics-based integration of hard sensor signatures (for example, radiological, chemical, and biological). Should more sensor data be acquired, the likelihood can be revised again with analogous mathematics.

The phrase “physics-based” emphasizes the fact that in many instances, different sensors share common components in their source-path-receiver function. The mathematical representation of these dependencies is key to optimal threat detection and identification. The mechanics of iteratively revising a threat likelihood can be quite transparent to frontline users. For example, the output to a Customs officer can be in the form of a ranking of possible threats that are consistent (or not) with intelligence data, hard sensor signatures, and situational awareness.

The sequential revision of threat likelihood is fundamental to Bayesian methods. Key aspects of Bayesian thinking include the utilization of information from the human sensor (for example, situational awareness), updating the human sensor signature with physical sensor data and possibly new intelligence, and clear mathematics to incorporate the consequences of actions into a decision. Within the context of homeland security, the fundamental Bayesian equation can be demonstrated as

P(Threat) x P(Data|Threat) = P(Threat|Data)

In words, this equation intuitively reads, “The prior-to-data probability of a threat revised with (times) the probability model of additional data gives (equals) the updated probability of a threat.” In 1988, Arnold Zellner³ demonstrated that under desirable assumptions about “information in equals information out,” the decision calculations must be based on Bayesian theory (conditional probability). Zellner asserted that Bayesian theory is 100% efficient. While this theory is quite sophisticated, this basic conditional probability equation is the foundation for all Bayesian decision techniques. It is mathematically combined with numerical representations of decision consequences, resulting in a framework that imposes standardization on information and careful thought on decision consequences. In total, Bayesian decision theory is the mathematics and philosophy dedicated to risk management.

There is research yet to be completed to facilitate the use of Bayesian decision frameworks in homeland security. The natural imposition of information standardization when using Bayesian thinking drives multiagency intelligence sharing in the form of likelihoods and probabilities. However, the development of mathematical summaries of data and information appropriate for use in a Bayesian framework requires research and established data standards. Bayesian thinking dictates that intelligence sharing is fundamentally more sophisticated than naively combining or linking databases and performing unguided data analysis on a monolithic database. At the front lines in the war on terrorism, research is necessary to effectively map the human sensor signature into the Bayesian equation, thus coupling situation awareness with intelligence and hard sensor signatures. This research should include calibration standards and assessments and regular, consistent training to fully utilize the power of the physical-human sensor tandem.

Building Interfaces Between Strategic and Technical Thinkers

As noted above, the natural imposition of information standardization when using Bayesian thinking facilitates multiagency intelligence sharing down to the front lines. This standardization requirement can also provide foundations for a road map of coordinated multiagency research—a bridge between strategic and technical thinkers. Strategic thinkers are concerned with

• threat detection
• consequences (human, economic, political) of false alarms
• clarity and traceability of decision processes
• legal and technical defensibility of homeland security missions
• logistics (for example, prioritization of research and operational resources)

and certainly many issues not listed.

Technical thinkers share the concerns of strategic thinkers, but for the most part focus on scientific and engineering research and development. Technical thinkers

• perform basic scientific research to improve chemical, radiological, and biological threat detection
• advance detection instrument capabilities (for example, advanced engineering leading to instrument miniaturization)
• advance communication capabilities
• develop sophisticated mathematics for data analysis, decision analysis, and instrument implementation

and many other research and development tasks directed at all aspects of homeland security.

Probably the most compelling point in favor of using Bayesian thinking in homeland security missions is the mathematical linkages between the strategic and technical thinker. To illustrate, the requirement that a threat be 99% detected and identified may be operationally achievable, but the associated costs for false alarms might make this operational requirement unrealistic. Also, a 5% chance of a missed threat in most scenarios is simply intolerable. With physics-based sensor integration, needed improvements can be identified because the source-path-receiver function is clearly and mathematically represented, thus providing the necessary information to create research and development roadmaps to improve threat detection capabilities (for example, to improve police and Customs officer training and advanced, molecule-selective transducer research).

There is nothing new here, except that a Bayesian decision framework can be used to clearly quantify the impact of each threat detection asset (human, physical sensor, intelligence) in an integrated decision capability. Statistical decisions depend greatly on local and regional threat detection capabilities mathematically coupled with decision consequences. Threat detection capability alone is insufficient for strategically optimal resource allocation. Thus, the Bayesian framework also provides clear insights into logistics issues.

The Research and Development Agenda for in-the-Field Sensor Integration

The Department of Defense Joint Directors of Laboratories Data Fusion Subpanel defined sensor fusion (integration) as “a multilevel, multifaceted process dealing with the automatic detection, association, correlation, estimation and combination of data and information for multiple sources.” It is not surprising that the Department of Defense has offered a fundamental definition of sensor integration, because sensor fusion in military applications is critical to battlefield operations. Sensor integration as used in homeland security certainly lies under this umbrella definition; however, the battlefield in homeland security is unique, as it is often fought within the borders of the 50 states, with each presenting unique challenges driven by geography, population distribution, differences in state governments and laws, and state economies. In-the-field decision algorithms for homeland security will

1. be based on the analysis of high-dimensional, multispectral signatures (massively large data signatures)
2. be a means of threat triage (decision evidence) in contrast to a crisp threat detection and identification
3. bind source-path-receiver physics and mathematics of disparate sensors into decision mathematics to
• identify barriers to operational requirements
• facilitate the mathematics of “tricorder”⁴ construction
• ameliorate calibration data requirements
• sharpen primary sensor signatures—for example, sharpening low-resolution nuclear spectra
4. require power-efficient algorithms to facilitate the use of low-power programmable integrated circuits
5. be grounded in decision mathematics appropriate for legal settings (United States and international)—that is, be lucid and technically rigorous
6. incorporate appropriate decision consequences at the field level (human and financial)
7. be appropriate to the demands of high-traffic triage—that is, not cumbersome to users

Summary

We advocate Bayesian statistical theory as the glue for homeland security mission components. Bayesian decision frameworks are expressly designed to manage risk. The fundamental Bayesian equation naturally integrates the human sensor, physical sensor data, and intelligence to arrive at a decision that is legally and technically defensible and lucid. Bayesian decision mathematics provides a mathematical link between the strategic and technical thinkers. This link leads to quantifiable insights into baseline operational performance and improvements and research requirements; as a consequence, it facilitates strategic decisions on research and development funding. Finally, a fully operational Bayesian framework provides optimal threat detection capability by lucidly gluing together all threat detection information, and it is fully adaptable to evolving threats.

References

Click on an end note number to return to the article.

1. “Bayesian logic is a branch of logic applied to decision making and inferential statistics that deals with probability inference: using the knowledge of prior events to predict future events”—TechTarget Network “Whatis?” See Jose Bernardo and Adrian F. M. Smith’s Bayesian Theory (New York: John Wiley & Sons, 2003).

2. Jay Davis and Don Proznitz, “Technical and Policy Issues of Counterterrorism—A Primer for Physicists,” Physics Today, April 2003.

3. Arnold Zellner, “Optimal Information Processing and Bayes’ Theorem,” American Statistician, vol. 42, no. 4, 1988, pp. 278–284.

4. A fictional device in Star Trek, used to scan an area for phenomena or to collect or transfer data, according to Wikipedia.