Is Better Nuclear Weapon Detection Capability Justified?

CREATE Research Archive Non-published Research Reports 6-1-009 Is Better Nuclear Weapon Detection Capability Justified? Niyazi Onur Bakir CREATE, nbakir@usc.edu Detlof von Winterfeldt University of Southern California, winterfe@usc.edu Follow this and additional works at: http://research.create.usc.edu/nonpublished_reports Recommended Citation Bakir, Niyazi Onur and von Winterfeldt, Detlof, "Is Better Nuclear Weapon Detection Capability Justified?" (009). Non-published Research Reports. Paper 8. http://research.create.usc.edu/nonpublished_reports/8 This Article is brought to you for free and open access by CREATE Research Archive. It has been accepted for inclusion in Non-published Research Reports by an authorized administrator of CREATE Research Archive. For more information, please contact gribben@usc.edu.

Is Better Nuclear Weapon Detection Capability Justified? Niyazi Onur Bakır and Detlof von Winterfeldt National Center for Risk and Economic Analysis of Terrorism Events University of Southern California, Los Angeles, CA 90089 Email: nbakir@usc.edu, Phone: (1) 81-917 June 1, 009 Abstract In this paper, we present a decision tree model for evaluation of the next generation radiation portal technology (Advanced Spectroscopic Portals or ASPs) to scan incoming containers non-intrusively against nuclear or radiological weapons. Advanced Spectroscopic Portals are compared against the current designs of portal monitors (plastic scintillators or PVTs). We consider four alternative deployment strategies: 1) Exclusive deployment of ASPs replacing all the PVTs currently deployed at U.S. ports of entry, ) Sequential deployment of ASPs with PVTs, ) Exclusive deployment of PVTs, ) Stop deployment of new portal monitors and continue inspections with the current capacity. i

The baseline solution recommends a hybrid strategy that supports the deployment of new and current designs of portal monitors together. However, this solution is found to be sensitive to the probability of attack attempt, the type of weapon shipped through ports of entry, the probability of successful detonation, detection probabilities and the extra deterrence that each alternative may provide. We also illustrate that the list of most significant parameters depends heavily on the dollar equivalent of overall consequences and the probability of attack attempt. For low probability and low consequence scenarios, false alarm related parameters are found to have more significance. Our extensive exploratory analysis shows that for most parametric combinations, continued deployment of portal monitors is recommended. Deployment of ASPs is optimal under high risk scenarios. However, we also show that if ASPs fail to improve detection capability, then extra benefits they offer in reducing false alarms may not justify their mass deployment. Keywords: decision trees; radiation portal monitors; advanced spectroscopic portals; border security; terrorism ii

Introduction Securing the United States (U.S.) homeland against a terrorist nuclear weapon delivered by various modes of transportation has been an increasing concern in recent years. Many homeland security experts contend that containers could be a perfect medium to deliver dangerous nuclear or radioactive material. In particular, a containerized nuclear or a radiological weapon strike has been considered a looming possibility in U.S. According to a survey of national security and non-proliferation experts compiled in 005 by U.S. Senator Richard G. Lugar (Lugar 005), the median value for the probability of a nuclear attack is 0.1 and 0. in the next 5 and 10 years respectively. For a radiological dispersion device (RDD, or aka dirty bomb), these figures jump to 0.5 and 0.0. While the question posed in this survey does not specify a geographical location, the responses are a clear testament to the shared concern among nation s top security experts. These concerns are not without substance. According to the International Atomic Energy Agency (IAEA), there were 87 confirmed cases of illicit trafficking in nuclear and radiological materials worldwide between 199 and 005 (see IAEA 006). A dirty bomb ingredient was discovered in more than 65% of these cases; whereas there were 16 confirmed incidents involving trafficking of highly enriched uranium (HEU) and plutonium (Pu). In a more recent case, British authorities foiled an attempt to smuggle an undisclosed amount of weapons-grade uranium from Russia to Iran and Sudan (see Townsend 007). Even a more disturbing report claimed that Al-Qaeda had already made an attempt to build a dirty bomb (see Gardner 00). Uncertainty about the security level of nuclear stockpiles in Russia, and growth in the number of states 1

belonging to the nuclear club only adds to the fears that terrorists may be able to build or acquire a nuclear or a radiological weapon. Many in the academic community concur with this conviction as well. Rosoff and von Winterfeldt (007) developed a model on the feasibility of a dirty bomb attack upon the twin Ports of Los Angeles and Long Beach. Their main conclusion was that if terrorists can successfully complete the planning and preparation tasks of an attack, there is a reasonable chance that they accomplish their mission. Some also believe that terrorists could build a crude nuclear weapon if they can acquire HEU or weapon-grade plutonium (see for instance Bunn and Wier 006, Martin 007, Allison 006, Cooper 00, Allison 00 and Maerli et al. 00). Unfortunately, the traditional methods of repelling a nuclear attack through missile defense systems do not shield against this rising threat. Because of this, since 00 the U.S. Department of Homeland Security (DHS) has been making an intensive effort to deploy non-intrusive inspection (NII) equipment to ports of entry. NII equipment currently operational at U.S. ports include gamma-ray and x-ray scanners, radiation portal monitors, personal radiation detectors and handheld radioactive isotope identification devices. Among these, radiation portal monitors are the only equipment to be used in primary inspections to detect the presence of nuclear or radioactive material. Earlier design of portal monitors, plastic scintillators (also known as poly-vinyl toluene, PVT, portal monitors) have been widely criticized for two main reasons. First, they are ineffective in detecting shielded nuclear material emitting low level of radiation (i.e. highly enriched uranium, HEU). Second, they cannot distinguish between harmless and dangerous nuclear and radiological materials. To address these concerns, the

Domestic Nuclear Detection Office (DNDO) under DHS contracted the development of advanced passive detection technology. This technology could be used during primary inspections and reduce the rate of false alarms while improving the detection performance against shielded nuclear and radioactive material. These next generation portal monitors, known as Advanced Spectroscopic Portals (ASPs), have already been through their initial engineering development phase. However, despite DHS s efforts to advocate the mass deployment of ASPs at U.S. ports of entry, Senate has not yet approved $1. billion requested in funding mainly due to concerns raised by the Government Accountability Office (GAO) and lawmakers (see GAO 007a). In this article, we present a decision tree model that examines the economic costs and benefits of deploying ASPs. We focus on a scenario of a single terrorist attack using either a nuclear weapon or a dirty bomb. We parameterize the probability of attack, the probability of detecting the weapon, and the effectiveness of ASPs in reducing the likelihood of a successful terrorist attack and providing more deterrence. The model accounts for the cost of false alarms, the economic consequences of a successful attack and the probability of discovering the weapon after smuggling through the border. While we do not obtain precise probabilities of critical events in the decision tree, we determine optimal decisions for various realistic parameter values in the model. We use Precision Tree software by Palisade Inc. to build the decision tree in an Excel spreadsheet enhanced by a user-friendly Visual Basic interface. The spreadsheet also includes a graphical display of the expected economic costs of an attack, the cost of false alarms and countermeasures (i.e. ASP vs. PVT). Sensitivity analyses are performed by Precision

Tree within the low and high end estimates of parameters modeling critical event probabilities and consequences. The rest of this paper is organized as follows. Section provides a detailed discussion of the decision tree including countermeasures, false alarms, and the parameter values associated with random events. In Section, we illustrate the results of our sensitivity analysis on system parameters. Section offers our concluding remarks. Figure 1 Decision Tree for Container Inspections at U.S. Ports ASP Attempt p A Nuclear q A RDD 1-q A HEU h A Plutonium 1-h A Detection r A No Detection Detection u A No Detection 1-u A Detection t A No Detection 1-t A Detonation s Detonation s No Detonation 1-s Detonation s No Detonation 1-s No Attempt 1-r A No Detonation 1-s 1-p A PVT/ASP Tree is repeated with exactly the same sequence of events. Nuclear PVT No RPMs Attempt p N q N RDD 1-q N HEU h N Plutonium 1-h N Detection r N No Detection Detection u N No Detection 1-u N Detection t N No Detection 1-t N Detonation s Detonation s No Detonation 1-s Detonation s No Detonation 1-s No Attempt 1-r N No Detonation 1-s 1-p N. Description of the Decision Tree Figure 1 displays part of the decision tree used in this study. The decision maker has four options: Deploy ASPs, Deploy PVTs and ASPs in sequence, Deploy PVTs and

Do nothing. The chain of events following each decision is identical. However, the probability and cost parameters associated with these events are different. The values of these parameters are assigned based on the estimates available in the open literature. The objective is to minimize overall expected equivalent costs (EEC) in dollar terms. EEC includes the cost of countermeasures, the dollar equivalent cost of human casualties, the impact on the U.S. economy and the cost of false alarms..1. Countermeasure Decisions DNDO has not completed its task to deploy portal monitors to U.S. ports. The current plan is to complete deployments by September 01 (GAO 007b). As of October, 008, 1,15 portal monitors have been installed (Medalia 008) and the goal is to bring that number up to,75 (GAO 008). We consider four alternatives that DNDO could consider in the future to reach this goal. First alternative, Deploy ASPs, considers designing an inspection scheme that is based on an exclusive use of ASPs for scanning containers. This includes both the replacement of PVTs that are currently operational at ports of entry and the installment of new ASPs to reach the goal by September 01. Second alternative, Deploy PVTs and ASPs in sequence, is based on DNDO s project plan to deploy new portal monitors to U.S. ports of entry. According to a recent report by the Government Accountability Office (GAO 008), a total of,75 portal monitors will be purchased under this plan. The goal is to use ASPs in both the primary and the secondary inspections at high volume ports and to keep PVTs in primary inspections at other ports. Third alternative, Deploy PVTs, calls for continued installment of PVTs without the use of ASP technology to meet the target by September 01. Finally, DNDO 5

has a Do nothing alternative which discontinues the deployment of new portal monitors and uses 1,15 PVTs that are currently installed for inspections. We primarily use the information provided in a recent report by the Government Accountability Office (GAO 008) to estimate the cost of each alternative. GAO 008 estimates procurement, deployment, maintenance, design and development as well as sustainment costs for the project plan that was prepared by DNDO in 006. DNDO s 006 plan is to deploy 1,0 ASPs and 1,58 PVTs for use in both primary and secondary inspections. We use the cost figures available in GAO (008) for the Deploy PVTs and ASPs in sequence alternative. The costs of other alternatives are calculated under the following assumptions. On average, each PVT costs $55,000 while each ASP costs $77,000 (GAO007a). The procurement cost as estimated in GAO (008) includes computer and spares procurement as well. Therefore, we use an approximate unit procurement cost of $100,000 for a PVT and $00,000 for an ASP machine. O Harrow (008) reports that the cost of deployment is $00,000 and $5,000 per unit for ASPs and PVTs respectively. This cost includes one year maintenance contract for each design. Annual maintenance cost is approximately $6,600 per PVT and $80,000 per ASP. Therefore we assume a unit deployment cost of $18,000 per unit of each portal monitor. Design and development cost should be zero for Do nothing and Deploy PVTs alternatives because GAO s estimate includes the cost of developing various designs of ASPs only. We assume that the personnel costs for security operations at U.S. ports of entry are the same under all strategies that call for deployment of new portal monitors. For the Do nothing alternative, we use a lower sustainment cost estimate of $50,000,000. 6

Table 1 Itemized 10-Year Cost of Countermeasure Alternatives, $ Procurement Deployment Maintenance Design & Development Sustainment Deploy ASPs, c A 1,00,078,08 818,91,176,08,17,915 6,77,57 6,988,656 Deploy PVTs & ASPs in seq., c PA 60,678,9 689,107,870 999,,69 6,77,57 6,988,656 Deploy PVTs, c P 150,5,771 78,,99 18,851,95 0 6,988,656 Do nothing, c N 0 0 8,19,77 0 50,000,000 Table 1 lists the estimates of total procurement, deployment, maintenance, sustainment and design and development costs under each alternative. The total cost for the Deploy ASPs alternative is calculated assuming that a total of,75 ASPs will be installed. Under the Deploy PVTs alternative, we assumed that a total of 1,609 PVTs will be installed, which is the difference between the number of portal monitors originally planned for deployment, and the number of portal monitors currently operational. The calculations were made based on a gradual deployment strategy and a % discount rate to account for the time value of money. The cost estimates obtained are rounded to the nearest million dollars. The baseline estimates and associated ranges for each alternative are listed in Table. Table Ranges for the Cost of Countermeasure Alternatives, (in million $) Min Base Max Deploy ASPs, c A,000,77 7,000 Deploy PVTs & ASPs in seq., c PA,600,100,800 Deploy PVTs, c P 500 1,177,000 Do nothing, c N 0 500 7

.. False Alarms While not explicitly included in the decision tree, the cost of false alarms under each alternative is calculated and factored into the decision. Table illustrates the base values and ranges of parameters related to the cost of false alarms. The most challenging piece in calculating the cost of false alarms is the estimation of the average cost a single false alarm, c f. To our knowledge, there is no study on the average cost of a false alarm in the open literature. Cost of a false alarm should include delay costs as well as the cost of physical examination. In a port security study by Center for Discrete Mathematics and Theoretical Computer Science (DIMACS), cost of physical cargo inspection is assumed to be $600 (DIMACS 006). Currently, Hong Kong Container Terminal Operators Association charges $1500-$000 (see Wein et. al. 007) per physical examination. In this paper, we assume that physical examination costs $1,500, and approximately 10% of containers that are flagged by false alarms are subject to further physical examination. Therefore, physical examination cost per container that triggers a false alarm is $150. Table Parameters Associated with the Cost of False Alarms Min Base Max Percentage of false alarms, f 1%.50% 10% Number of incoming containers per year (in millions), N 15 0. 5 Average cost of a false alarm, $, c f 0 150 1000 Percentage of containers scanned by portal monitors, s RPM 98 98 100 False alarms may cause delays as well. In Martonosi et al. (00), an hour delay cost is assumed to be $6 for regular and $60 for time sensitive load. Currently, most 8

false alarms are usually resolved quickly (Rooney 005). Therefore, delay costs are assumed to be negligible. We believe a range between $0 and $1000 for the average cost of a single false alarm is reasonable and a base value of $150 is justified if in practice most false alarms are usually not causing delays. The rate of false alarms is an important piece in estimating the cost of false alarms. Under the current scheme, it has been reported that about.5% of scanned containers generate false alarms (Rooney 005). The primary benefit of ASPs is the potential reduction in the rate of false alarms. Cochran (008) states the ASPs the have the potential to reduce false alarms regardless of whether they are deployed at primary inspections replacing PVTs or at secondary inspections complementing PVTs in primary inspections. According to an Associated Press report, officials believe that ASPs should reduce false alarms by 90% (see Sullivan 008). We take this as the baseline estimate for the false alarm benefit of the DNDO s 006 project plan. As such, the alternative to Deploy PVTs and ASPs in sequence reduces false alarms by 90%. On the other hand, the alternative to Deploy ASPs is assumed to perform slightly better and reduce false alarms by 95%. The parameters modeling the effectiveness of each countermeasure alternative to reduce the cost of false alarms are listed in Table. Table False Alarm Effectiveness Parameters, % Reduction Min Base Max Reduction in the percentage of false alarms for the Deploy PVTs and ASPs in sequence alternative, fe PA 0 90 100 Reduction in the percentage of false alarms for the 0 95 100 Deploy ASPs alternative, fe A 9

The number of incoming containers per year, N, is another parameter of interest. Approximately 9. million containers arrive at seaports whereas 11. million incoming trucks cross land ports of entry each year (Bonner 005 and Wasem et. al. 00). As such, the base value for N is assumed to be 0. million containers per year. Currently, 98% of incoming containers are scanned by the portal monitors (see Winkowski 008). Therefore, we set the base value of the percentage of containers scanned to 98%... Sequence of Events Each countermeasure decision is followed by the same sequence of events, albeit with different probabilities of occurrence. For notational convenience we use the same lower case letter for the probability of the same event under each decision alternative, but a different subscript is used to highlight the impact of the decision. For instance, u is used for the probability of detecting HEU with a subscript that highlights which alternative is chosen for inspections (i.e., u A for Deploy ASPs, u PA for Deploy PVTs and ASPs in sequence, u P for Deploy PVTs and u N for Do nothing ). The first event of interest is the probability of attempt for an attack using a containerized radiological or nuclear weapon through ports of entry, p. We use the median probability estimates for a nuclear and a radiological weapon attack in Lugar (005) as an anchor. Since they do not identify a particular geographical location or a particular mode of an attack, these estimates have to be refined for our purposes. In this paper, we consider both a nuclear and a radiological dispersion device (RDD) attack. Assuming both events are independent, 10

P (A nuclear or an RDD attack attempt) = P (A nuclear attack) + P (An RDD attack) P (A nuclear and an RDD attack) = 0.0 + 0.0 0.08 = 0.5. Then we calculate the probability of attack attempt, p using the formula below: p = P(A nuclear or an RDD attack) x P(U.S. target A nuclear or an RDD attack) x P(Weapon smuggled through a port of entry A nuclear or an RDD attack) = p W x p US x p POE. (1) We assume that U.S. will be targeted with probability 0.5 (i.e., p US = 0.5). An attack using a nuclear weapon or an RDD could be carried out by either acquiring sources in U.S. or building the weapon in a foreign country and bringing it in between ports of entry. However, many ports are close to critical infrastructure and urban areas, so they can be attractive targets for an attack. Besides, one could argue that acquisition of nuclear material is quite difficult in U.S. and stealing radiological material from a facility in U.S. may require more intensive preparation than a terrorist would need to do in a foreign country. As such, we use a base value of 0.75 for p POE. Therefore, using (1), p = p W x p US x p POE = 0.5 x 0.50 x 0.75 ~ 0.0. () We do not have any credible open source information that could suggest us to use a range other than between 0 and 1 for any of the conditional probability values in (). In other words, we use an unrestricted range between 0 and 1 for p. 11

The base value of 0.0 for p does not incorporate any extra deterrence that deployment of new radiation portal monitors may offer. Deployment of new PVTs and ASPs may be another deterrent to terrorists. We revise the probability of attempt under each alternative by multiplying p with a deterrence coefficient (i.e., For Deploy ASPs, p A = p x (1-d A ) ). Table 5 lists deterrence coefficients under each alternative that requires deployment of new radiation portal monitors. Note that Do nothing alternative does not provide an extra deterrent (i.e., d N = 0). The deterrence levels are chosen to reflect improvements in the probability of detecting HEU under different alternatives (see Table 6). Furthermore, since the detection improvement of these alternatives over the Do nothing alternative is minimal, the baseline extra deterrence levels are set to values ranging between 0.10 and 0.0 (i.e., 10% to 0% deterrence). Table 5 Extra Deterrence of Countermeasure Alternatives Min Base Max Extra deterrence of "Deploy ASPs", d A 0 0.0 1 Extra deterrence of Deploy PVTs and ASPs in sequence, d PA 0 0.15 1 Extra deterrence of Deploy PVTs, d P 0 0.10 1 The next node in our tree is regarding the choice between an RDD and a nuclear weapon given an attempt. The probability of events branching from this node does not depend on the particular countermeasure alternative chosen. We assume that terrorists will choose to attack using a nuclear weapon with probability 0.5 (π) and using an RDD with probability 0.75 given an attempt. If a nuclear weapon is chosen, the source of nuclear material used has an influence on the probability of detection at a U.S. port of entry. Therefore, the node that follows the nuclear weapon event represents the choice 1

between plutonium and HEU as a source of nuclear material. Both events are assumed to be equally likely irrespective of the countermeasure alternative. Table 6 Detection Probability under Various Countermeasure Alternatives "Deploy ASPs" Min Base Max Probability of detecting an RDD, r A 0 0.90 1 Probability of detecting a plutonium-based nuclear weapon, t A 0 0.90 1 Probability of detecting an HEU-based nuclear weapon, u A 0 0.5 1 "Deploy PVTs and ASPs in sequence" Min Base Max Probability of detecting an RDD, r PA 0 0.90 1 Probability of detecting a plutonium-based nuclear weapon, t PA 0 0.90 1 Probability of detecting an HEU-based nuclear weapon, u PA 0 0.5 1 "Deploy PVTs" Min Base Max Probability of detecting an RDD, r P 0 0.90 1 Probability of detecting a plutonium-based nuclear weapon, t P 0 0.90 1 Probability of detecting an HEU-based nuclear weapon, u P 0 0.0 1 "Do nothing" Min Base Max Probability of detecting an RDD, r N 0 0.85 1 Probability of detecting a plutonium-based nuclear weapon, t N 0 0.85 1 Probability of detecting an HEU-based nuclear weapon, u N 0 0.5 1 After the weapon choice is clarified, the question is whether the weapon arriving at a port of entry is interdicted by non-intrusive inspections. Detection probability depends on the material used to build the weapon and the particular countermeasure alternative chosen. Table 6 illustrates the base values and ranges used for detection probabilities. Irrespective of the countermeasure alternative chosen, the ranges used for the probability of detecting the threat material are the same. Both PVTs and ASPs have the capability to detect gamma radiation which is likely to be emitted from an RDD and neutron radiation which is emitted by a plutonium-based nuclear weapon. On the other hand, detection of HEU is quite difficult due to the low dose of radiation emitted. 1

Terrorists may also shield the weapon to reduce the probability of detection. Therefore, the probability of detecting an HEU-based nuclear weapon is considerably lower for each countermeasure alternative. Furthermore, experts believe that deployment of ASPs will not improve HEU detection capabilities significantly (see Cochran and McKinzie 008, Cochran 008). The target is an urban area that will be hit after the weapon crosses the border. It is important to smuggle the weapon through the border. However, there is still a chance of failure. Law enforcement agencies may foil the plot before a successful execution, or it may not be possible to detonate the weapon. Because of this, we add a final node at the end to model the detonation event. The probability of successful detonation depends on the type of the weapon, but the countermeasure alternative chosen has no impact on it. We use a base probability of 0.75 for an RDD (s RDD ) and 0.5 for a nuclear weapon (s N ). The ranges for both of these parameters are unrestricted... Economic Consequences The economic consequences of a successful attack include overall cost to the U.S. economy and the dollar equivalent of fatalities. Overall cost to the U.S. economy includes direct and indirect consequences of business disruption, evacuation costs, property damage values and decontamination costs. The cost of fatalities is computed by multiplying the value of life, which we set to $5 million in the base case, with the number of fatalities. The value of life is allowed to fluctuate between 0 and $10 million. Estimation of the number of fatalities is trickier. For an RDD attack, the number of fatalities depends on the location of attack, the source and the amount of radiological 1

material used as well as the weather conditions that influence the path of the plume. We do not elaborate on the mechanics of the radiation exposure. Instead we refer to high scenario estimates used in Rosoff and von Winterfeldt (007) for an RDD attack. As such, we assume that the number of fatalities due to blast and acute radiation effects as well as latent cancers will have a base value of 100 with a range between 0 and 00. Clearly, fatality figures for a nuclear attack are going to be more dramatic. The estimates may vary widely depending on the location of the attack, the size of the bomb and how the attack is executed. A nuclear attack claimed an estimated number between 90,000 and 10,000 in Hiroshima whereas the range of estimate is between 60,000 and 80,000 in Nagasaki 1. It is estimated that as high as 1 million people may die in a nuclear attack targeting Manhattan (see Abt Associates 00). However, according to most nuclear security experts, it is unlikely that terrorists will acquire a nuclear weapon that has the yield to generate such a huge number of fatalities. For instance, according to a study published by Council on Foreign Relations, setting off a one-kiloton nuclear weapon in Manhattan would generate about 00,000 casualties (Garwin 00). In this regard, we assume that a nuclear attack will generate casualties between 50,000 and 00,000 with a base value of 150,000. The estimates of overall costs to the U.S. economy vary. Gordon et al. (005) studies the impact of an RDD attack upon the Ports of Los Angeles & Long Beach and finds that an attack followed by a 10 day shutdown could cost $ billion to the U.S. economy. However, this excludes important pieces such as decontamination and property damage costs which could be significant for an urban area attack. According to 1 According to Radiation Effects Research Foundation (RERF). Available at http://www.rerf.or.jp/general/qa_e/qa1.html 15

an ongoing study by Defence Research and Development Canada, an urban area attack could cost up to $50 billion in Toronto, $80 billion in Vancouver and $75 billion in Windsor (Bronskill and Bailey 007). Our base value for economic costs of an RDD attack, E RDD is $00 billion (with a range between $100 billion and $600 billion). The cost of a nuclear weapon attack should be much higher. A RAND study estimates the cost of nuclear attack at the Ports of Los Angeles & Long Beach to be around $980 billion excluding decontamination and business disruption costs (see Meade and Molander 006). Business disruption and decontamination costs could easily add another $1 trillion to this estimate. Therefore, we use a base value of $ trillion for the overall cost of a nuclear attack in an urban area, E N, and assume that E N fluctuates between $500 billion and $ trillion.. Discussion of the Results Our model consists of 5 parameters and alternative decisions. The optimal decision is the one that minimizes overall expected equivalent costs (EEC) for a given combination of parametric values. We built the model as an Excel spreadsheet that involves a user friendly Visual Basic interface and a decision tree was added using Precision Tree software of Palisade Inc. Using base case values discussed in Section, the model recommends Deploy PVTs and ASPs in sequence alternative with an EEC of approximately $17 billion and offers Deploy PVTs as the second best. Do nothing alternative performs worse than the other three (EEC ~ $1.6 billion), which suggests that the Domestic Nuclear Detection Office (DNDO) should continue the deployment of portal monitors to improve containerized cargo security. 16

The tornado diagram including all the parameters shows that EEC is sensitive to 9 parameters. Figure compares the impact of the top 0. We observe that the probability of attempt is the most significant parameter. Two crucial parameters that determine the likelihood of a successful nuclear attack are also quite significant. In general, since the negative consequences of a nuclear attack are dramatically higher, the tornado diagram highlights the sensitivity of EEC to parameters associated with a nuclear attack. Not surprisingly, all deterrence parameters are quite influential on EEC due to their direct impact on the probability of attempt. Figure Tornado Diagram to Analyze the Sensitivity of EEC to the Top 0 Significant Parameters in the Model when p = 0.0 Probability of an attack attempt Probability of successful detonation nuclear Probability of nuclear attack attack Extra deterrence of "Deploy PVTs" Extra deterrence of "Deploy ASPs" Probability that an HEU source is used nuclear attack Extra deterrence of "Deploy PVTs and ASPs in sequence" Probability of detection under "Deploy PVTs" HEU Probability of detection under "Deploy PVTs and ASPs in sequence" HEU Economic consequences of a nuclear attack (in million $s) Probability of detection under "Do nothing" HEU Probability of detection under "Deploy ASPs" HEU Casualties after a nuclear attack Dollar value per casualty (in million $s) Economic consequences of an RDD attack (in million $s) Probability of successful detonation RDD Probability of detection under "Deploy PVTs and ASPs in sequence" RDD Probability of detection under "Deploy PVTs" RDD Probability of detection under "Deploy ASPs" RDD Probability of detection under "Deploy PVTs and ASPs in sequence" Plutonium -50% -150% -50% 50% 150% 50% 50% % Change from Base EEC Value 17

As far as the weapon material is concerned, detection of HEU under various alternatives is distinguished as critical. None of the current technologies is assumed to identify HEU with a high probability. Therefore, further reduction of terrorism risk by correctly identifying a containerized HEU-based nuclear weapon is recommended by our model. We also observe that detection probabilities under the base case decision are more significant than under other alternatives. Table 7 Parameter Base Value Changes for Scenarios in Figure Low Consequence Scenario Economic Consequences of a Nuclear Attack (in million $s) 1,000,000 Economic Consequences of an RDD Attack (in million $s) 100,000 Casualties after a Nuclear Attack 50,000 Casualties after an RDD Attack 50 Low Probability Scenario Probability of Attack Attempt 0.05 The most significant false alarm related parameter is the average cost of a false alarm. While the model does not discount the importance of false alarms, the consideration of extreme consequences in this study causes the parameters associated with the risk of terrorism rank higher than the parameters that determine the cost of false alarms. Therefore, the percentage of false alarms and the average cost of a false alarm are not among the most significant 0 parameters. Figure, which illustrates top 18 significant parameters for low consequence and low probability scenarios, supports this claim. Parameters associated with the frequency and the cost of false alarms become more significant as the risk of terrorism is reduced. Furthermore, some deterrence parameters lose significance in low probability and low consequence scenarios. Parameters that measure the economic consequences of an attack retain their 18

significance in the low consequence scenario. Parameters measuring human losses are found to be more significant in low consequence and low probability scenarios as well. Figure Tornado Diagram to Analyze the Sensitivity of EEC to the Top 18 Significant Parameters in the Model under different scenarios (see Table 7) a Low Consequence Scenario Probability of an attack attempt Probability of successful detonation nuclear Probability of nuclear attack attack Economic consequences of a nuclear attack (in million $s) Casualties after a nuclear attack Probability that an HEU source is used nuclear attack Probability of detection under "Deploy PVTs" HEU Economic consequences of an RDD attack (in million $s) Extra deterrence of "Deploy ASPs" Probability of detection under "Deploy ASPs" HEU Average cost of a false alarm Percentage of false alarms Extra deterrence of "Deploy PVTs" Extra deterrence of "Deploy PVTs and ASPs in sequence" Probability of detection under "Do nothing" HEU Probability of detection under "Deploy PVTs and ASPs in sequence" HEU Dollar value per casualty (in million $s) Probability of detection under "Deploy PVTs" RDD -50% -50% -150% -50% 50% 150% 50% % Change from Base EEC Value b Low Probability Scenario Probability of an attack attempt Probability of successful detonation nuclear Percentage of false alarms Dollar value per casualty (in million $s) Economic consequences of an RDD attack (in million $s) Probability of successful detonation RDD Probability of nuclear attack attack Probability that an HEU source is used nuclear attack Extra deterrence of "Deploy PVTs" Probability of detection under "Deploy PVTs" HEU Economic consequences of a nuclear attack (in million $s) Probability of detection under "Do nothing" HEU Extra deterrence of "Deploy PVTs and ASPs in sequence" Casualties after a nuclear attack Average cost of a false alarm Probability of detection under "Deploy PVTs and ASPs in sequence" HEU Probability of detection under "Deploy PVTs" RDD Cost of "Deploy PVTs" (in million $s) -800% -00% 0% 00% 800% 100% % Change from Base EEC Value Decision sensitivity analyses for the most significant parameters reveal a tradeoff between all the alternatives. The two way sensitivity plot in Figure analyzes the optimal decision as the probability of an attack attempt (p) and the probability of successful detonation of a nuclear bomb (s N ) are perturbed. A non-linear trade-off between alternatives exists as a function of these parameters. The plot suggests that as 19

the threat of a nuclear attack increases, the model recommends deployment of the new technology. In fact, as long as p exceeds 0.05 and s N is over 0.0, deployment of ASPs is optimal for most parametric combinations. However, exclusive deployment of ASPs is not recommended for any parameter value unless p exceeds 0.16 and s N is over 0.08. The current status is the best alternative if the probability of an attempt is less than 0.0. Figure Two way Sensitivity Analysis for Parameters p and s N Figure 5 is the two way sensitivity plot between s N and the probability of a nuclear attack given an attempt is made for an attack (π). A similar picture emerges except a shift in the region in which Do nothing alternative is optimal. This shift is primarily due to the reduced likelihood of a successful attack. Note that the model considers both a nuclear and an RDD attack. When π is high, the probability of an RDD 0

attack is low. In Figure 5, the region in which Do nothing is optimal represents parametric combinations under which terrorists likely weapon of choice is almost sure to fail. Therefore, improving the current inspections is not recommended. The plot also suggests that deployment of ASPs is not recommended when either s N or π is below 0.06. Furthermore, the baseline decision is close of the breakeven point for Deploy PVTs rather than Deploy ASPs in Figures and and 5. Significant changes to the baseline parameter values are required to justify exclusive deployment of ASPs rather than the hybrid deployment alternative as considered by DNDO. Figure 5 Two way Sensitivity Analysis for Parameters s N and π Relatively speaking, false alarms do not have much influence on EEC. However, they have some influence on the optimal decision because the baseline case suggests 1

that EEC values for the top three alternatives are close. This is illustrated in the two way sensitivity plot for the percentage of false alarms (f) and the average cost of a false alarm (c f ). Deploy ASPs may be justified for reducing the cost of false alarms only when the false alarm rate exceeds 8% and the cost of a false alarm is over $800. For most parameter values, the hybrid deployment alternative is economically sound. However, under baseline assumptions, if f is below % or c f is less than $100, the optimal decision switches to the exclusive deployment of PVTs. Figure 6 Two way Sensitivity Analysis for Parameters c f and f

Economic consequences of a nuclear attack (in million $s) Extra deterrence of Deploy PVTs Extra deterrence of Deploy PVTs Extra deterrence of Deploy PVTs Extra deterrence of Deploy PVTs Figure 7 Exploratory Analysis for Parameters p, E N, d PA, and d P $ trillion $ trillion $1 trillion 50% 0% 10% 50% 0% 10% 50% 0% 10% 1 Do nothing Deploy PVTs Deploy PVTs and ASPs in sequence Deploy ASPs d PA 55% d PA 5% d PA 15% d PA : Extra deterrence of Deploy PVTs and ASPs in sequence $500 billion 50% 0% 10% 0.05 0.0 0.0 Probability of an attack attempt The exploratory analysis that plots the sensitivity of EEC as a function of p, the economic consequences of a nuclear attack (E N ), the deterrence provided by Deploy PVTs (d P ) and the deterrence provided by Deploy PVTs and ASPs in sequence (d PA ) is given in Figure 7. Each circle in small rectangular regions indicates the optimal decision for a given parameter combination. Perhaps surprisingly, the economic consequences of a nuclear attack have low influence on the optimal decision. This influence erodes as the probability of an attack attempt increases. On the other hand, the deterrence parameters have significant impact on both EEC and the optimal decision. If p is low, then the hybrid deployment is preferred only when the economic consequences are high, and it provides to 5 times higher deterrence in comparison to exclusive

Probability of detection under Deploy PVTs and ASPs in sequence HEU Probability of detection under Deploy PVTs HEU Probability of detection under Deploy PVTs HEU Probability of detection under Deploy PVTs HEU Probability of detection under Deploy PVTs HEU deployment of PVTs. As p increases, Deploy PVTs and ASPs in sequence become optimal even when the difference between deterrence levels is either low or zero. Figure 8 Exploratory Analysis for Parameters p, u P, u PA, and u A 0.55 0.5 0.5 0.5 0.0 0.5 0.5 0.0 0.5 0.5 0.0 0.5 1 Do nothing Deploy PVTs Deploy PVTs and ASPs in sequence Deploy ASPs u A 0.55 u A 0.5 u A 0.5 u A : Probability of detection under Deploy ASPs HEU 0.5 0.5 0.0 0.5 0.05 0.0 0.0 Probability of an attack attempt Parameters associated with HEU detection are quite significant in the model as well. Figure 8 is a plot investigating the impact of HEU detection under deployment alternatives and p. Not surprisingly, low values of p favor the Deploy PVTs alternative. However, as p increases, even small increases in the HEU detection capability lead to a change in the optimal decision. The level of sensitivity of the optimal decision to other detection parameters is relatively low. This suggests that the

HEU detection capability of the new technology will be a main determinant of whether its mass deployment is an economically sound decision. Table 8 Parameter Base Value Changes for the Equal Detection Probability Scenario "Deploy ASPs" Min Base Max Extra deterrence, d A 0 0.10 1 Probability of detecting an HEU-based nuclear weapon, u A 0 0. 1 "Deploy PVTs and ASPs in sequence" Min Base Max Extra deterrence, d PA 0 0.10 1 Probability of detecting an HEU-based nuclear weapon, u PA 0 0. 1 "Deploy PVTs" Min Base Max Extra deterrence, d P 0 0.10 1 Probability of detecting an HEU-based nuclear weapon, u P 0 0. 1 "Do nothing" Min Base Max Probability of detecting an HEU-based nuclear weapon, u N 0 0.0 1 A major criticism by the Government Accountability Office has been based on the initial field results that suggest that ASPs do not perform better than PVTs in detecting nuclear material. In this paper, we assumed that the deployment of ASPs increases the probability of detecting a HEU-based nuclear weapon by 0.05. Parallel to this assumption, we assumed that both Deploy PVTs and ASPs in sequence and Deploy ASPs alternatives should provide an extra deterrence because the probability of interdiction is assumed to be better. Our last analysis is to see the behavior of the optimal decision if we drop the assumption that ASPs perform slightly better than PVTs in detecting HEU and thus is a higher deterrent to the adversary. Table 8 lists the parametric values used under this equal detection probability scenario. In this case, the 5

base case optimal alternative is Deploy PVTs. The behavior of the optimal decision becomes totally independent of parameters that collectively determine the probability of a successful attack and its consequence. Figure 6 Two way Sensitivity Analysis for Parameters c f and f under Equal Detection Probability Scenario Figure 9 shows that there is fine trade-off between Deploy PVTs and ASPs in sequence and Deploy PVTs alternatives when we drop the assumption that ASPs improve the probability of detecting HEU. In fact, the trade-off is observed when the false alarm rate exceeds 1% and the cost of a false alarm is over $100. Economic justification of Deploy PVTs and ASPs in sequence alternative is still possible under 6

this scenario particularly when the cost of false alarms is high. For example, if the percentage of false alarms is % and the cost of a false alarm is twice the baseline value assumed in this paper, the hybrid deployment alternative is cost effective.. Conclusions Improvement of inspections to interdict nuclear or radiological weapons hidden in incoming containers at U.S. ports of entry has been a pressing concern for the Department of Homeland Security (DHS) since 9/11. In an effort to reduce the risk, DHS has been deploying portal monitors that inspect containerized cargo nonintrusively. However, current designs of portal monitors (PVTs) have been criticized for their limited capability in detecting shielded highly enriched uranium (HEU) and for generating high rate of false alarms. In response, DHS is planning to install the next generation portal monitors (ASPs) to improve the non-intrusive inspections. In this study, we have developed a decision tree model that evaluates alternative policies on possible replacement of PVTs by ASPs. The base case decision recommends design of an inspection scheme that employs PVTs and ASPs together. The analysis also indicates that the objective function value is quite sensitive to system parameter values. The probability of attack, the probability of successful detonation of nuclear weapon, and the probability that a nuclear weapon is used in the attack are the most important parameters. The tornado diagram generated around base values reveals that the deterrence parameters, the probability of detecting HEU under various alternatives and the consequences of a nuclear attack are among the top 0 significant parameters. 7

The price tag for the Deploy ASPs alternative is considerably higher than that of all other alternatives. Therefore, it is the optimal decision only if there is a fairly high probability of success to terrorists. Deploy PVTs is economically justified if the average cost of false alarms is below $100 or the percentage of false alarms is below 1%. The baseline decision is close to the breakeven point between Deploy PVTs and Deploy PVTs and ASPs in sequence alternatives. Slight changes in the probability of an attack, and baseline HEU detection probability could result in changes to the baseline optimal decision. Parameters associated with false alarms which are relegated to be of secondary importance around the base value. However, they are critical in determining the optimal deployment strategy under low probability and low consequence scenarios. They are also significant if we assume that ASPs do not offer any extra utility in detecting nuclear material. The results indicate in this case that the cost and the rate of false alarms could still justify deployment ASPs along with PVTs. As a result, keeping all other parameters fixed at their base values, Deploy PVTs and ASPs in sequence alternative is recommended over Deploy PVTs alternative if the cost of false alarms exceeds $00 and PVTs generate % or higher rate of false alarms. The analysis strongly discourages replacement of all PVTs with ASPs at ports of entry if ASPs are not also proven to improve detection capabilities. The model presented in this paper includes 5 parameters. We assign reasonable ranges based on unclassified information and analyze whether replacement of current portal monitors with the next generation technology is justified. The results indicate that an inspection scheme that employs a hybrid deployment strategy is reasonable, and 8

exclusive deployment of the new technology requires further proof that they improve the detection capability as well. References Abt Associates. 00. The Economic Impact of Nuclear Terrorist Attacks on Freight Transport Systems in an Age of Seaport Vulnerability. Allison, G. 00. Nuclear Terrorism. Times Books, New York, NY. Allison, G. 006. The Will to Prevent. Harvard International Review. 8-50-55. Bonner, R. C. 005. Statement of Robert C. Bonner Hearing before the permanent Subcommittee on Investigations, Senate Committee on Homeland Security and Governmental Affairs. Available at http://hsgac.senate.gov/_files/stmtbonnercbp.pdf. Bronskill, J., S. Bailey. 007. Dirty Bomb Would Cause Panic, Cost Billions: Study. TheStar.com. http://www.thestar.com/article/1670. Bunn, M., A. Wier. 006. Terrorist Nuclear Weapon Construction: How Difficult?. The Annals of the American Academy of Political and Social Science. 607 1-19. 9

Cochran, T. B. 008. The Utility of Advanced Spectroscopic Portal Monitors for Interdicting WMD. Statement of Dr. Thomas B. Cochran before the Committee on Homeland Security and Governmental Affairs United States Senate, Washington, D.C. Available at http://hsgac.senate.gov/public/_files/09508cochran.pdf. Cochran, T. B., M. G. McKinzie. 008. Detecting Nuclear Smuggling. Scientific American. 98-98-10. Cooper, M. H. 00. Nuclear Proliferation and Terrorism. CQ Researcher. 1-1 97-0. DIMACS Center. 006. Port of Entry Annual Report. GAO (Government Accountability Office). 007a. Combating Nuclear Smuggling: DHS s Cost-Benefit Analysis to Support the Purchase of New Radiation Detection Portal Monitors Was Not Based on Available Performance Data and Did Not Fully Evaluate All the Monitors Costs and Benefits. U.S. Government Accountability Office, GAO-07-1R, Washington D.C. GAO (Government Accountability Office). 007b. Combating Nuclear Smuggling: DHS s Decision to Procure and Deploy the Next Generation of Radiation Detection Equipment Is Not Supported by Its Cost-Benefit Analysis. U.S. Government Accountability Office, GAO-07-581T, Washington D.C. 0

GAO (Government Accountability Office). 008. Combating Nuclear Smuggling: DHS s Program to Procure and Deploy Advanced Radiation Detection Portal Monitors Is Likely to Exceed the Department s Previous Cost Estimates.U.S. Government Accountability Office, GAO-08-1108R, Washington D.C. Gardner, F. 00. Al Qaeda was making dirty bomb. BBC News Online. http://news.bbc.co.uk//hi/uk_news/71165.stm. Garwin, R. L. 00. Nuclear and Biological Megaterrorism. Available online at http://www.fas.org/rlg/0081-terrorism.htm Gordon, P., J.E. Moore, H. W. Richardson, Q. Pan. 005. The Economic Impact of a Terrorist Attack on the Twin Ports of Los Angeles-Long Beach. In The Economic Impacts of Terrorist Attacks, edited by H. W. Richardson, P. Gordon and J.E. Moore II. Edward Elgar, Northampton, MA IAEA (International Atomic Energy Agency). 006. Illicit Trafficking and Other Unauthorized Activities Involving Nuclear and Radioactive Materials. IAEA 005 Fact Sheet. Lugar, R.G. 005. The Lugar Survey on Proliferation Threats and Responses. http://lugar.senate.gov/reports/npsurvey.pdf. 1