3

Technical MDV Capabilities and Research and Development

This section of the report assesses current technical monitoring, detection, and verification (MDV) capabilities and identifies opportunities for new or expanded research and development (R&D) efforts to improve capabilities.¹ This assessment focuses on key and emerging MDV capabilities in three topical focus areas—the nuclear fuel cycle, nuclear test explosions, and arms control—as well as open-source assets and data and advanced data analytics, which cross-cut the topical areas. Figure 3-1 illustrates the role that these topical areas play across the nuclear weapon acquisition process.² Note that the early steps of the nuclear weapon acquisition process involve motivation, capability development, and proliferation intent; these concepts are described in Box 3-1.

The committee chose to approach this section of the report by topical area because these different missions are often separated in the policy and operational communities. However, common technologies support MDV in each of these areas. For example, radiation detection is applied to MDV for the fuel cycle, nuclear test explosions, and arms control, although the exact application may differ (see Appendix L for a detailed table showing the cross-cutting nature of radiation detection and other technologies). While the MDV policy and operational communities tend to differentiate the topical mission areas, the MDV technical community is primarily organized around these technical categories. A scientist

___________________

¹ Since the committee was unable to fully assess some components of the technical MDV landscape for this interim report, the committee has not recommended how these identified R&D opportunities should be prioritized. The question of R&D priorities will be revisited in the final report.

² Due to time constraints, the committee did not significantly address MDV technologies for smuggled nuclear materials or weapons, weaponization activities, or the export of sensitive technology and expertise for this interim report.

at a national laboratory who studies atmospheric transport, for example, is likely to apply their knowledge to both nuclear fuel cycle and nuclear explosion MDV, while a scientist studying tags and seals may consider both safeguards and arms control applications of these technologies.

3.1 RESEARCH AND DEVELOPMENT PRIORITIES

This section discusses the MDV technical priorities of the interagency as well as the National Nuclear Security Administration (NNSA), the primary MDV R&D organization. At this time, the committee has not made an assessment of the balance of R&D priorities across the MDV mission but will revisit this question in the final report.

3.1.1 Interagency R&D Priority Guidance

The Nuclear Defense Research and Development (NDRD) Strategic Plan for Fiscal Years (FY) 2020–2024 (NSTC, 2019) outlines interagency policy objectives and R&D priorities for five different elements of nuclear defense. Two of these elements, (1) nonproliferation and verification, and (2) detection and attribution, are relevant to this study. The policy objectives within these elements related to the MDV mission are to

• predict, detect, and characterize nuclear proliferation and weapons development activities;
• provide timely and reliable verification and monitoring of states’ compliance with nuclear safeguards and arms control agreements; and
• maintain effective operational capabilities to detect and characterize nuclear materials worldwide.

Specific R&D priorities associated with each of these policy objectives are shown in Appendix M. The NDRD Strategic Plan neither ranks these priorities nor notes which are already being pursued and where gaps remain. As discussed in Section 2.1.3, this limits the utility of the strategic plan as an interagency guiding document.

3.1.2 NNSA Priorities and Budget

Agencies have their own individual priorities that ideally are aligned with the national guidance. The current top priorities of the NNSA Office of Defense Nuclear Nonproliferation (NNSA/DNN R&D), which is the largest MDV R&D funder, are

• delivery of space-based nuclear detonation detection sensors to the Air Force;
• proliferation detection research with an emphasis on uranium conversion and enrichment, as well as weaponization; and
• stewardship of technical competencies in the national laboratories (DNN R&D, briefing to committee, September 21, 2020).

Table 3-1 shows NNSA MDV funding from FY2017–2020, broken down by topic area (NNSA, communication to committee, September 18, 2020). Funding over these four years underwent a roughly linear rise (23 percent), but not all topic areas saw an increase. The Monitoring and Verification Field-Testing Program (i.e., test beds; see Section 2.2.2 for additional information) saw the most significant growth with a 50 percent increase in funding (2.9 percent increase in portfolio share). This rapid growth indicates the high priority of the test beds.

The Space-Based Detection portfolio has also increased with the overall NNSA funding growth (29 percent, 1.2 percent increase in portfolio share). This portfolio is large because DNN R&D conducts expensive high technology

TABLE 3-1 NNSA Integrated Priority List for MDV, FY2017–2020

Topical Area	FY17 ($K)	FY18 ($K)	FY19(SK)	FY20 ($K)	FY17-FY20 Change	FY17-FY20 Change in PF Share
University Program	16,000	17,000	17,000	17,000	6%	-0.5%
Nuclear Weapons Development and Material Production	58,500	70,520	70,520	70,380	20%	-0.4%
Nuclear Weapons and Material Security	43,400	35,400	35,400	34,800	-20%	-3.5%
Monitoring and Verification Field-Testing	61,000	72,078	82,078	91,280	50%	2.9%
Space-based Detection	132,267	163,605	163,605	170,322	29%	1.2%
Ground-based Detection	12,000	8,590	8,590	9,712	-19%	-1.0%
Nonproliferation Enabling Capabilities (Near-field, Remote, Data Science)	33,655	38,202	38,202	38,000	13%	-0.7%
Nonproliferation Stewardship Program	0	0	0	22,480	N/A	4.1%
International Nuclear Safeguards*	52,429	54,313	52,429	55,962	7%	-1.7%
Nuclear Verification*	28,773	33,482	32,273	33,208	15%	-0.5%
Total*	438,024	493,190	500,097	543,144	24%	N/A

* As detailed in Congressional Budget documents, not all of this funding is directed to technology development projects.

readiness level (TRL)³ research, development, testing, and evaluation (RDT&E) in this area to deliver satellite sensors to the Air Force.

All other topical areas lost portfolio share to these two efforts and the Nonproliferation Stewardship Program, which was established in FY2019 (see Section 2.2.1 for additional details on this program). Despite the current policy focus on future arms control treaties, MDV R&D for arms control technologies has not been a recent priority, with flat or decreasing budgets in both the DNN Nonproliferation and Arms Control (NPAC) Nuclear Verification portfolio and the DNN R&D Nuclear Weapons and Material Security portfolio. This is not due to a lack of research ideas; DNN R&D noted that they have funded a smaller percentage of received proposals on arms control technology development compared to other topics (DNN R&D, communication to committee, September 29, 2020).

Another NNSA portfolio with notably flat funding over the past three years is the Enabling Capabilities Portfolio in DNN R&D, which includes R&D on near-field and remote detection as well as data science topics. However, DNN R&D noted that from 2017 to 2020, the data science component of this portfolio has increased by approximately 50 percent, while radiation detection R&D has decreased in priority over the past decade (ibid.).

3.2 MDV FOR THE NUCLEAR FUEL CYCLE

MDV for the nuclear fuel cycle involves looking for signatures indicative of each step (Figure 3-2): uranium mining and processing, conversion, enrichment, fuel fabrication, energy generation/isotope production in a reactor, reprocessing, and spent fuel/waste storage.⁴ The ability to monitor and detect each of these steps is important for detecting proliferation, but the steps associated with undeclared uranium enrichment and undeclared plutonium production (irradiation and separation) are of primary focus. While most attention has rightfully centered on uranium and plutonium, less common fissile materials, such as neptunium, have also been assessed for proliferation risk.

Nuclear fuel cycle MDV is conducted both cooperatively and unilaterally. Unilateral MDV for the nuclear fuel cycle will be discussed in more detail in the final report.

Cooperative fuel cycle MDV is primarily carried out by the International Atomic Energy Agency (IAEA), which was already implementing safeguards in some states when it was assigned the responsibility for establishing a comprehensive safeguards system to verify the Nuclear Non-Proliferation Treaty (NPT). As of June 2020, 175 states had a comprehensive safeguards agreement (CSA), which is required for all non-nuclear weapons states signatories under the NPT. The IAEA’s role under a CSA is to verify that a state’s declared nuclear materials

___________________

³ See 1570 for a description of TRLs.

⁴ For more information on the nuclear fuel cycle, see IAEA (2011a) and DOE, (n.d).

are not diverted to nuclear weapons or other nuclear explosive devices. However, as illustrated by the discovery of an undeclared Iraqi nuclear weapons program in the aftermath of the 1991 Gulf War, the IAEA safeguards system was unable to ensure that undeclared activities were not pursued by a state with a CSA. In response to this revelation, the IAEA in 1997 adopted the Model Additional Protocol (AP), which gives expanded access to inspectors and seeks to understand the totality of a state’s declared and potentially undeclared nuclear activities (IAEA, 1997). More than 135 states have signed an AP.

This section addresses fuel cycle MDV that can be conducted on-site and in the surrounding environment. Stand-off sensing techniques like overhead sensing and seismic monitoring can also support fuel cycle MDV. Commercial overhead sensing is addressed in Section 3.5.1; government capabilities in these areas will be addressed in the final report.

3.2.1 MDV for Declared Facilities

The IAEA conducts on-site MDV in declared facilities under safeguards. The IAEA is well-equipped for its safeguards mission, but there are still opportunities to improve its capabilities. As described in Section 2.3.1, the United States conducts R&D to support the IAEA through DNN R&D, NPAC, and

the International Safeguards Project Office. These organizations develop tools and methods to ensure timely detection of diverted material and to deter/detect undeclared activities. Some of the R&D needs associated with four important classes of tools and methods—nuclear material accountancy, containment and surveillance (C/S), unattended monitoring, and environmental sampling—are outlined below.⁵

Nuclear material accountancy involves determining the amount of nuclear material present within a defined area to assess consistency with a state’s declarations. Tools and methods to conduct nuclear material accountancy include item counting, non-destructive analyses, and destructive analyses on small amounts of sampled material. One need that the IAEA identified in this area is improvement of non-destructive analysis methodologies for verification of uranium at bulk facilities (IAEA Dept. of Safeguards, briefing to committee, September 14, 2020). Other R&D needs identified by the committee include measurement methods for alternative fuel cycles (e.g., thorium)⁶ and better nuclear data to improve measurements and uncertainty quantification.⁷

C/S methods and tools are used to maintain chain of custody and continuity of knowledge, monitor activities, and record indications of undeclared activity or containment breach. C/S tools include cameras, portal monitors, and tags and seals, which are described in Box 3-2. Tags and seals have many applications in safeguards, such as tracking uranium hexafluoride (UF)₆ cylinders.⁸ (Tags and seals can also be used in arms control MDV applications, as discussed in Section 3.4.)

Tags and seals are ideally inexpensive, quick for IAEA inspectors to install, difficult or impossible for the inspectors to attach incorrectly, and simple to examine during on-site visits. Research needs for C/S capabilities include improved unique identification for tags and seals to reduce risk of counterfeiting/spoofing, improved tamper indicating enclosures (volumetric seals), and application of data science methods to improve event detection (see Section 3.5).

Unattended monitoring technologies perform continuous qualitative/quantitative measurements of nuclear processes or activities without requiring the presence of an inspector. Unattended monitoring provides continuity of knowledge between IAEA inspections and reduces demands on IAEA inspectors. The latter is becoming more and more important as the IAEA faces an increasing volume of verification activities (IAEA Dept. of Safeguards, briefing to committee,

___________________

⁵ For more information on these classes of techniques, see IAEA (2011a).

⁶ DNN R&D recently conducted a scoping study on safeguards R&D needs for the thorium fuel cycle (DNN R&D, briefing to committee, October 16, 2020). See also Worrall et al. (2016).

⁷ Nuclear data include fission yields, cross-sections, and half-lives. The experiments to measure these data can be expensive and require special facilities. DNN R&D has recently funded R&D to improve nuclear data for non-destructive assay measurements and modeling measurements and modeling (DNN R&D, briefing to committee, October 16, 2020).

⁸ NNSA recently conducted R&D to develop standard tagging methods for UF₆ cylinders which will facilitate tracking and accountability (SEAB, 2020).

September 14, 2020). These technologies have been a significant recent focus of NNSA. For example, NNSA developed the On-Line Enrichment Monitor (OLEM; see Box 2-3) which has been deployed as part of the MDV regime for the Joint Comprehensive Plan of Action (JCPOA) (Fournier and Gaspar, 2016). Some challenges for unattended monitoring technologies include the need for increased battery life and improved reliability for sustained operation in a variety of environments. Testing unattended processing monitoring systems can be challenging in the United States because of the diminishing commercial nuclear activities. As noted in Box 2-3, the OLEM underwent a field trial at a URENCO facility in the Netherlands. A U.S. test bed could potentially help with development and testing of these capabilities.

Environmental sampling was added to the IAEA’s toolbox in the 1990s and is now a key capability of the agency (Box 3-3). The IAEA collects swipe samples inside and immediately outside enrichment plants and other facilities to help confirm the absence of undeclared nuclear material or nuclear activities. Swipe samples are used to determine characteristics such as uranium enrichment levels and the actinide content of reprocessing waste streams, which can help confirm whether a state’s activities are consistent with its safeguards declarations. Swipe samples are analyzed at the IAEA Environmental Sampling Laboratory and at laboratories in the Network of Analytical Laboratories (IAEA, 2012). Research priorities communicated by the IAEA for environmental sample analysis include the development of new methodologies to determine the age of collected samples and analyze impurities relevant to the origin of the source materials (ibid.).

In each of these categories of technologies, it is essential that NNSA engage with the IAEA to understand IAEA needs and work to ensure the IAEA has more than one technical option for critical safeguards tools. The IAEA also underscored that ensuring long-term sustainability of assets is a current challenge (IAEA Dept. of Safeguards, briefing to committee, September 14, 2020).

In addition to improving current tools and methods, there are opportunities to strengthen and expand the safeguards regime in order to better detect early proliferation.⁹ A few opportunities for expanded focus are outlined below.

Other Fuel Cycle Steps

IAEA safeguards traditionally begin mid-way through the conversion process¹⁰ and focus primarily on the enrichment, fuel fabrication, reactor, and reprocessing steps of the fuel cycle.¹¹ This is logical since these steps involve the most attractive material from a proliferation point of view (HEU and Pu).

However, monitoring the initial steps of the fuel cycle, mining and processing, can improve the ability to detect undeclared activities early. While mining and processing operations are not monitored under a CSA, the AP provides for expanded declarations and access to assess these operations. Monitoring uranium mines and mills will require new MDV methods that monitor ore quality and track and weigh shipments. The IAEA is currently monitoring Iran’s uranium mining sector under the JCPOA; this experience may provide valuable insight into how uranium mining can be better monitored globally.

On the other end on the fuel cycle, there is work to be done in safeguarding spent fuel in storage. Spent fuel stored in casks can only be verified by verifying the seal on the cask lid; no alternative method exists. Once spent fuel is placed in a geological repository for long-term storage, as is being pursued in Europe, beyond monitoring of entrances and exits, there will be no way to access it and technically verify that it is still there and has not been removed.

Ongoing efforts in NNSA are addressing the need for direct measurement of spent fuel (DNN R&D, briefing to committee, October 16, 2020), but additional methods are needed to ensure that the contents of spent fuel casks have not changed and that all original items are still present. NNSA should support continued R&D to develop capabilities to obtain a direct measurement of the plutonium and uranium content of spent fuel and monitor and verify that spent fuel transfers from reactor to storage locations are as declared. This may include

___________________

⁹ Preparing for new types of facilities is a top priority R&D need identified by the IAEA in the 2018 R&D Plan (IAEA, 2018).

¹⁰ UF₆ cylinders are weighed, tagged, and sealed before they leave the conversion plant, and checked and weighed again upon arrival at an enrichment plant.

¹¹ See Table 1 in IAEA (2011a) for a summary of the main safeguards techniques deployed in each of these types of facilities.

collaborations with commercial cask manufacturers to incorporate safeguards into the design to enable monitored storage.

Emerging Reactor Types, Alternative Fuels, and New Enrichment Methods

To maintain the effectiveness of the safeguards regime, MDV capabilities must keep pace with technological advancements including alternative fuels, alternative methods, and new reactor types. The United States cannot dismiss non-traditional materials and methods solely because they are not economically viable compared to currently commercially available technologies. These materials and methods may be pursued for R&D in the capability development phase shown in Figure 3-1. As noted in Section 1.3, Iraq built calutrons for enrichment not because they were efficient or economical but because the knowledge and technology were available (Erkman et al., 2008).

Recent R&D on enrichment monitoring and detection capabilities have focused primarily on gaseous centrifuges. While this work is important, it is also critical to develop methods to monitor and detect emerging or non-traditional enrichment methods such as laser isotope separation (e.g., SILEX).¹²

R&D on reprocessing safeguards tools and processes have similarly focused on commercially viable Purex and Urex techniques. There are several other reprocessing techniques that have been demonstrated at the R&D or bench scale, from which a small scale reprocessing facility can operate to produce fissionable material.¹³

There is also a need to focus on alternative fuel materials like thorium or fuel types like liquids using molten salts and pebbles using pyrolytic graphite.¹⁴ Most of the tools and methods used to inspect fresh fuel and detect missing rods in an assembly before it is placed in a reactor are built for uranium fuel types and will not be suitable for alternative fuels (Kovacic et al., 2018).

Other emerging safeguarding challenges identified by the IAEA (IAEA, 2017, 2020a) and external experts (Rockwood et al., 2018) include advanced nuclear reactors, transportable nuclear power plants, accelerator-driven systems, and additive manufacturing. New MDV capabilities will be needed in the future to address these areas.

___________________

¹² The Separation of Isotopes by Laser Excitation (SILEX) technology was developed in Australia in the 1990s. Recent plans to commercialize the SILEX technology in the United States have faltered (WLN, 2018).

¹³ Reprocessing techniques that have been demonstrated on a small scale include Chemex, Redox, and amine extraction. Note that in 1977, researchers at Oak Ridge National Laboratory designed a quick and simple reprocessing system (Ferguson, 1977).

¹⁴ Pebble bed modular reactors and molten salt reactors do not use identifiable fuel bundles and have to be treated as bulk facilities. DNN R&D is currently exploring safeguards measurements of molten salt reactors under development by the United States (DNN R&D, briefing to committee, October 16, 2020).

Small-Scale Research Reactors

Small research reactors with low thermal power ratings (typically less than 15 MWth) have not been a high safeguards priority and are inspected less frequently since it would take more than one year to produce a significant quantity¹⁵ of plutonium. However, a proliferator may only wish to acquire enough material to develop one or two weapons and deem research reactors to be a viable proliferation pathway to do so. In addition, research reactors can be part of the proliferation preparation phase shown in Figure 3-1 and may be a key enabling infrastructure and proliferation-capability builder. Recovery of the small amounts of plutonium produced in target materials could help develop small scale reprocessing capabilities. Once recovered, the plutonium would be available for metallurgical studies to help design and fabrication planning. These types of activities build the expertise and knowledge base to support proliferation.

Current safeguards practices generally focus on the core fuel and not on the experiments or the target materials used. These target materials can be frequently changed and/or removed between inspection periods. New MDV techniques are needed to provide improved awareness of the experimental activities conducted, the target materials being used, and the methods for concealing these types of activities.

3.2.2 Environmental MDV for Undeclared Facilities

Some nuclear activities, such as enrichment, conversion, and reprocessing, can emit chemical effluents (often at trace levels) into the environment.¹⁶ These effluents contain information about the process or activity from which they originated. Some proliferation-related effluents have relatively long environmental lifetimes and thus can provide insights into historical as well as contemporary activities.

As described in the previous section, the IAEA currently conducts swipe and environmental sampling inside and immediately outside declared facilities as part of its regular safeguards inspections. In this scenario, the collectors know what effluent characteristics are expected based on provided facility operational information and are looking for any anomalous data that would suggest that a state’s declarations are not correct or complete.

The use of environmental sampling to monitor for and detect unknown (or undeclared) activity is quite different from on-site/near-site environmental

___________________

¹⁵ A significant quantity (SQ), defined by the IAEA, is the “approximate amount of nuclear material for which the possibility of manufacturing a nuclear explosive device cannot be excluded” (IAEA, 2011a). An SQ of HEU is defined as 25 kg, while an SQ of plutonium is defined as 8 kg.

¹⁶ Enrichment and conversion may release trace amounts of UF₆ which is quickly converted into uranyl fluoride (UO₂F₂) in the presence of atmospheric water. Reprocessing may release trace amounts of ⁸⁵Kr. For more information on chemical effluents, see Kemp (2016) and Schoeppner (2018).

sampling of declared facilities, because the geographic relationship between the sample collection location and the unknown activity is itself unknown. The general environmental sampling process is illustrated in Figure 3-3. If there is no known or suspected source site, a wide area or region is sampled and inverse transport calculations are necessary to link detected signatures to a source location. This generates significant technical uncertainties related to sampling statistics, detection thresholds, and modeling uncertainties.

If other information provides a cue for the possible existence of an unknown activity in a certain region, the problem is easier since there are bounds to the sample collection required and a known link to the source location. However, identifying sampling areas still requires an accurate prediction of the aquatic and/or atmospheric transport of the effluent away from the source, as well as the fate of the effluent in the environment.¹⁷ Once a region of interest is identified, sampling may be aimed at atmospheric particulates, gases, aquatic regimes involving water sampling, sediment collection, and/or biota collection in the appropriate areas depending on the region in question.

The IAEA has explored the application of wide-area environmental sampling (WAES) to “assist the Agency in drawing conclusions about the absence of undeclared nuclear material or nuclear activities over a wide area” (IAEA, 1997).¹⁸

___________________

¹⁷ Effluents may be modified by the environment through chemical reactions with environmental constituents such as water, oxygen, and hydrocarbons; through agglomeration with other atmospheric contaminants/constituents; and/or through radioactive decay.

¹⁸ The Model Additional Protocol defines WAES as “the collection of environmental samples (e.g., air, water, vegetation, soil, smears) at a set of locations specified by the Agency for the purpose of assisting the Agency to draw conclusions about the absence of undeclared nuclear material or nuclear activities over a wide area” (IAEA, 1997).

The general idea for WAES is not to target a specific facility or geographical location but instead to monitor for undeclared activities over larger regions that could be up to thousands of square kilometers. If implemented, WAES would enhance the IAEA’s ability to detect undeclared activities. However, such a large sampling program would be expensive to implement, and WAES was previously assessed to be cost prohibitive (IAEA, 1999). In addition, implementing WAES as part of the global safeguards regime is politically challenging since effluents can provide technical insights into nuclear processes and thus potentially reveal activities, or information about activities, that neighboring states would not want disclosed. States may also deny access to some sample collection locations to diminish the chances of collecting effluents of interest.

While some of these challenges are political and outside the control of the R&D enterprise, the R&D enterprise can continue to conduct research to improve the viability of WAES and environmental monitoring in general. Given the high cost and effort associated with WAES, there is particular interest in improving sampling and analysis speed and efficiency with the added resources of emerging machine learning (ML) and data analytic techniques, which are addressed in Section 3.5. Other areas of needed R&D are described below. Many of these R&D topics are also necessary to improve radionuclide (RN) monitoring for nuclear test explosion detection, which is described further in Section 3.3.4.

Source Term Release Mechanisms

Releases may be episodic or continuous and may interact with the facility or environmental media through which they transit during and after release (e.g., filters or geologic materials) or evolve during the release (e.g., radioactive decay, chemical forms). Characterizing release mechanisms, including possible holdups and transformation, can help in interpreting detected effluents. As new types of reactors and other fuel cycle technologies are developed and deployed, the characteristics of their source terms must be studied as well.

Environmental Fate

Effluent releases may be modified by the environment through chemical reactions with environmental constituents (e.g., water, oxygen), through agglomeration with other atmospheric contaminants/constituents, and/or through radioactive decay. The fate of various effluents is an important area in need of additional research to improve detection capabilities. In particular, research is needed to assess how the evolution of the chemical forms of various effluents of interest depend on various environmental factors such as humidity, temperature, and biota affinity. Furthermore, additional research is needed to understand how environmental fate may change over time as radioactive decay produces daughter species that may have entirely different chemical behaviors, and consequently different reactions in the environment.

Atmospheric¹⁹and Aquatic Transport

In addition to potentially being modified by the environment, effluent releases are transported away from release site through atmospheric and/or aquatic mediums. Relating a sample to its source location is therefore non-trivial. Integrated models in both atmospheric and aquatic regimes are needed to better identify where to sample in the case of known source locations and help attribute detections to possible source locations via backtracking when source locations are not known. Increased focus on aquatic modeling is necessary if it is to be used within the environmental-sampling construct.²⁰ Research is needed to link mesoscale and microscale atmospheric and aquatic models²¹ and to improve resolution and uncertainty quantification. In developing such a complex modeling system, algorithms should be verified and validated against experimental data.

Unidentified Source Location

Considering all sample collection locations as a network rather than single collection sites and coupling the aggregated detects and non-detects with the improved atmospheric and aquatic transport models mentioned above can greatly improve the ability to backtrack anomalous detections to possible source locations. Backtracking from a single location is dependent upon the accuracy of the transport models and the weather data that are used in the models. The possible regions for source location can quickly become large, especially if the transport distance from source to collection locations is large. Considering time resolved collections from multiple locations can significantly improve the ability to backtrack to a source location and reduce the estimated size of the region that contained the source. Associated, capability-enhancing research includes developments in fast, three-dimensional linked mesoscale and microscale transport models that include dilution and error quantification, in-situ analyses of samples at the various collection sites to eliminate transport-to-laboratory time, improved minimum detectable limits, and anticipatory cuing of one sampler location based on detects at other sampler locations coupled with transport predictions (see Section 3.5 for more details on anticipatory cuing of sensors).

___________________

¹⁹ Two atmospheric transport modeling capabilities are the Department of Energy (DOE) National Atmospheric Release Advisory Center (NARAC) operated by Lawrence Livermore National Laboratory (LLNL) and the Air Force Technical Applications Center (AFTAC). AFTAC focuses on atmospheric transport for nuclear test monitoring (see Section 3.3). NARAC is a national resource response center for emergency response to any and all accidental or intentional atmospheric releases of nuclear, radiological, chemical, biological, and hazardous natural materials.

²⁰ The advantage of an aquatic program is that effluent often binds with colloid and particle material in the water and is deposited in the sediment, where it can be measured at a later time.

²¹ A desired effort by the Air Force, the Rapid Environmental Modeling Unified System, could improve transport modeling capabilities and positively impact the ability to successfully collect samples and interpret the measurement results (AFTAC, briefing to committee, September 23, 2020).

Background Characterization²²

Conducting a WAES effort relies on the sampling and analysis of large numbers of individual samples from a region in search of signatures relevant to undeclared activities. Having continuous sampling data to establish the background as a function of time can aid in discriminating the signal from the natural or acceptable man-made signatures measured. The RN background has latitudinal variations, north/south hemispherical variations, weather induced variations, and local variations due to proximity to known sources such as reprocessing facilities and medical isotope production facilities. Understanding the background for nuclides of interest is an ongoing, continuous requirement to enable environmental monitoring and is becoming increasingly challenging as increased amounts of nuclear effluents are released from contemporary nuclear activities. Additional research is necessary and could be a key confidence building measure. The development of the unidentified source location capability discussed above would also aid the characterization of atmospheric backgrounds. Background mapping is an area where open-source data could be particularly valuable.

Field Testing

In addition to advancing these MDV capabilities, conducting a field test for atmospheric and aquatic measurements would illustrate the utility, costs, and operational problems to be solved for environmental sampling within the context of a useful proliferation monitoring tool and confidence building measure. Such tests might illustrate to the various state parties how WAES would be implemented and thus decrease their concerns.

3.2.3 Findings and Recommendations

Finding 7. Fuel cycle MDV technologies must evolve to keep pace with the expanding universe of nuclear activities, in terms of both emerging technologies and growth in the number of nuclear activities.

1. IAEA resources have remained constant for a number of years despite increasing MDV demands, implying future MDV may be less comprehensive and less frequent unless more efficient and effective MDV techniques are developed.
2. Current MDV technologies and methods were developed to detect traditional uranium-fueled reactors, gaseous centrifuge enrichment plants, and reprocessing facilities. MDV technologies for emerging reactor designs, alternative enrichment techniques, alternative fuels, and small scale, non-traditional approaches to reprocessing need development support.

___________________

²² Background characterization is also a key research need for nuclear test explosion detection.

3. Current MDV paradigms focus on validating declarations, deterring illicit material diversions, and detecting unknown, undeclared activities. Expanding the MDV paradigm to include motivation and early capability development may enhance opportunities to dissuade and/or counter proliferation behavior and encourage responsible, peaceful use of nuclear energy and technology.

Recommendation 7. NNSA should prioritize R&D efforts that (a) enhance efficiency, ease of use/deployment, and sustainability of safeguards tools and technologies; (b) address MDV for advanced reactors, non-traditional and emerging enrichment techniques, and small and/or non-traditional reprocessing technologies; and (c) enhance capabilities to monitor and detect early capability development that could be a potential proliferation threat.

Finding 8. Understanding and modeling source term mechanisms, the environmental fate, and atmospheric/aquatic transport of proliferation effluents are key to identifying when and where to sample and gaining insight into proliferation activities from analyzed samples. New analytic approaches that concurrently consider results from multiple sampler locations coupled with atmospheric and aquatic transport models can improve the identification of potential source locations.

Recommendation 8. DNN R&D, in coordination with interagency partners, should continue to support R&D to improve understanding of and develop more accurate models for source terms, environmental fate, and atmospheric/aquatic transport. Field tests should be conducted to assess limitations of the models. These efforts will enhance MDV capabilities for both the nuclear fuel cycle and nuclear test explosions (see Section 3.3 below) and should include the following:

1. Developing models of effluent release processes and mechanisms from both fuel cycle processes (including new and emerging reactor and fuel cycle technologies) and underground nuclear explosions.
2. Developing linked mesoscale and microscale models for atmospheric and aquatic modeling of effluents of interest.
3. Clarifying the effect of temperature, humidity, UV light, and other pertinent environmental factors on effluent species to determine the nature and rate of physical and chemical changes.
4. Developing integrated analytic processes to analyze environmental sampling results from all relevant sampling locations as a network, coupling their temporally resolved results with atmospheric and aquatic transport models can improve plume source location capability.

Finding 9. To enable the application of WAES as a proliferation and nuclear explosion MDV tool, additional work is needed to characterize known sources of radionuclides and regional background variations.

Recommendation 9. DNN R&D, in collaboration with interagency and international partners, should support R&D to characterize known sources of radionuclides of interest and regional background variations to enhance MDV capabilities for both the nuclear fuel cycle and nuclear test explosions (see Section 3.3 below).

3.3 MDV FOR NUCLEAR WEAPONS TEST EXPLOSIONS

Over the course of the Cold War, the United States and Soviet Union conducted more than 1,700 nuclear tests as they developed new nuclear weapons capabilities (ACA, 2020). While these tests were initially conducted atmospherically, several factors, including public pressure to reduce radioactive fallout and the desire to better hide the results of a test from an adversary, led states to seek to limit tests in the atmosphere and move nuclear testing underground. In the late 1950s, the United States accepted a Soviet testing moratorium in advance of an eventual partial ban on nuclear explosive tests. Negotiations on an atmospheric test ban stalled for several years due to the difficulty of verifying such an agreement. In 1963, the parties ultimately agreed to a treaty²³ banning nuclear testing in the atmosphere, under water, and in space (but not underground), relying on national technical means (NTM)²⁴ for monitoring and verifying the treaty. The United States and the Soviet Union subsequently negotiated a 1974 Threshold Test Ban Treaty limiting the size of underground tests to the equivalent of 150,000 tons of TNT, which entered into force December 11, 1990.²⁵

Throughout this period, states worked to design the parameters of and negotiate a Comprehensive Nuclear-Test-Ban Treaty (CTBT) banning all nuclear weapon test explosions or any other nuclear explosion. The CTBT opened for signature in 1996. While the CTBT has been signed and ratified by almost every country, it has not yet been ratified by the United States and the other remaining countries needed for it to enter into force.²⁶ The treaty text stipulates the construction of multiple types of sensors to monitor the world for nuclear explosions.

___________________

²³ Limited or Partial Test Ban Treaty.

²⁴ National technical means refers to unilateral MDV capabilities.

²⁵ Ratification of the treaty was delayed due to disagreements on how the yield limit would be verified. To address this challenge, in 1988, the United States and Soviet Union conducted the Joint Verification Experiment, a collaboration to measure the explosive yields of nuclear tests by each side (Robinson, 2016).

²⁶ The United States, Egypt, Iran, Israel, and China have not ratified the Comprehensive Nuclear-Test-Ban Treaty Organization, and North Korea, India, and Pakistan have not signed.

Current U.S. policy (DoD, 2018) states that while the United States will not seek to ratify the CTBT, it will continue to support the architecture established to monitor and verify the treaty, described below in Section 3.3.1.

Nuclear test explosion MDV capabilities were previously assessed in the 2012 National Academies report The Comprehensive Nuclear-Test-Ban Treaty: Technical Issues for the United States. In this section, the committee focuses on advances made since 2012, particularly in seismic and RN monitoring.²⁷ The committee also notes key challenges, including the accuracy of nuclear explosion yield estimates, detecting very low-yield tests, and detecting tests conducted using evasive methods such as decoupling the test in a large underground cavity.²⁸

3.3.1 Comprehensive Nuclear-Test-Ban Treaty Organization International Monitoring System

Article IV of the CTBT calls for a global verification regime to monitor compliance with the treaty. The Preparatory Commission for the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO) was established to construct and operate the network of sensors detailed in the treaty and to develop an onsite inspection protocol.

Significant progress has been made in recent years toward full deployment of the International Monitoring System (IMS). More than 300 stations have been completed, representing an infrastructure investment of about $1.2 billion (the United States has contributed about 20 percent).²⁹Figure 3-4 shows the completion percentage of the various types of IMS stations as of November 2020.

At the core of the IMS is an array of seismic sensors, “the most effective technology for monitoring underground nuclear-explosions testing” (NASEM, 2012, p. 43).³⁰ The network of seismic sensors detects the propagation of elastic waves through the earth. Seismologists have developed reliable techniques to distinguish the seismic waves produced by the hundreds of natural earthquakes that occur every day from the signals that are produced by an underground explosion. The 2012 National Academies CTBT report found that the threshold for CTBTO

___________________

²⁷ The committee focused on seismic and RN monitoring because they are the primary monitoring technologies for underground tests (NASEM, 2012).

²⁸ The 2012 National Academies CTBT report notes, “Constraints placed on nuclear-explosion testing by the monitoring capabilities of the IMS, and the better capabilities of the U.S. NTM, will reduce the likelihood of successful clandestine nuclear-explosion testing, and inhibit the development of new types of strategic nuclear weapons” (p. 2). In addition, “countries with less nuclear-explosion testing experience would face serious costs, practical difficulties in implementation, and uncertainties in how effectively a test could be concealed” (p. 112).

²⁹ This cost includes establishment of the International Data Centre (IDC) Division and On-Site Inspection Division as well.

³⁰ As shown in Figure 3-4, the IMS has two seismic networks. The primary network sends data continuously in real time to the IDC. The auxiliary network, which takes advantage of existing seismic stations, only sends data to the IDC upon request (CTBTO, 2021b).

IMS seismic detection for fully coupled nuclear explosions is below 1,000 tons of TNT equivalent worldwide (NASEM, 2012),³¹ and advancements have been made in seismic monitoring since 2012 (see Section 3.3.3 below).

In addition to the seismic network, the IMS uses arrays of hydroacoustic sensors (to detect sound waves produced by explosions underwater or near water) and infrasound sensors (to detect sound waves produced by explosions in the atmosphere) and a system of RN sensors (to detect the fission and activation products produced by a nuclear explosion which may leak from an underground test) to monitor compliance with the treaty.

The CTBTO International Data Centre (IDC) receives information on a continuous basis from IMS stations and sensors.³² These data are also provided continuously to states that have arrangements to receive the data, including the United States.³³ Some IMS data are also available upon request to the MDV research community and researchers in other fields.

___________________

³¹ The 2012 National Academies CTBT report also notes that the CTBTO IMS “provides valuable data to the United States, both as an augmentation to the U.S. NTM system and as common baseline for international assessment and discussion of potential violation when the United States does not wish to share NTM data” (NASEM, 2012, pp. 38–39).

³² The IDC relies on the Global Communications Infrastructure (GCI), based on nine Hughes geostationary satellites. The satellites provide data from the stations to the IDC and on to National Data Centers (NDCs) at about 15-20 Gbytes/day (5.5 Tbytes/yr). The system has met the entry-into-force requirement that 85 percent of the network data be available through the GCI. The annual operation cost is $30 million per year. The IDC is currently leveraging work being done to modernize the U.S. NDC, since the systems have shared capabilities.

³³ AFTAC is the U.S. NDC for the CTBTO and receives all IMS data.

3.3.2 U.S. Atomic Energy Detection System

The United States also monitors the world for nuclear test explosions via the U.S. Atomic Energy Detection System (USAEDS), which is operated by AFTAC and includes atmosphere and space, geophysics, and nuclear materials components. The atmosphere and space event detection capabilities are supported by U.S. Nuclear Detonation Detection System satellite constellations that monitor for nuclear explosions at varying altitudes using optical flash, radio frequency (RF), and neutron/gamma/X-ray detectors. Underground, at sea, and low altitude event detection capabilities are supported by seismic, hydroacoustic, and infrasonic sensors of the USAEDS network, supplemented by corresponding data from the IMS, including ground-based RN data. AFTAC also has ready access to data from open networks, such as the U.S. Geological Survey (USGS)/National Science Foundation (NSF) Global Seismic Network (GSN) and other national and open-source regional seismic networks (see Section 3.5.1). Atmospheric gas and particle sample collection is performed using ground-based air samplers and aircraft.³⁴ Together with IMS RN data, the samples are processed by an extensive laboratory network in the United States.

3.3.3 Advances in Seismic Monitoring

Monitoring nuclear test explosions has become increasingly challenging as sequential testing treaties have moved testing underground, limited the size of underground tests, and then sought to ban all tests outright. The CTBT has significantly reduced the size of all explosion and non-explosion events that must be detected, located, and identified on a global basis. This has motivated development of techniques that can be used for small events, which are best recorded at regional distances (< 1600 km). Such techniques supplement classical MDV methodologies, intended to detect events larger than magnitude 4 (approximately 1 kiloton yield) that are recorded at larger, teleseismic distances.

Small nuclear events are difficult to distinguish from background events and variation. Accounting for the tens of thousands of annually detected small background events has prompted development of advanced data analysis, including artificial intelligence (AI) for rapid event location and identification. Many AI procedures being developed for earthquake monitoring applications with large regional seismic arrays may provide capabilities that can be used in seismic MDV operations. This includes cross-correlation methods that exploit vast accumulated archives of waveforms for previously identified sources, which can quickly screen for repeat quarry blasts and closely located seismicity.

Substantial progress in regional wave analysis for both large and small tests has been made since the 2012 National Academies CTBT report was released. Improved remote MDV capabilities have been demonstrated by many analyses of

___________________

³⁴ See Box 1-2 for more information on the WC-135 aircraft.

signals from the North Korean nuclear tests of 2006, 2009, 2013, 2016 (twice), and 2017 (Walter and Wen, 2018). Nuclear test explosion detection and identification has proved robust for the North Korean tests, which included an event that was likely sub-kiloton. However, the accuracy of nuclear yield estimation appears to remain at about the traditional “factor-of-two” level for an uncalibrated test site based on the scatter in yield estimates. This is due to uncertainties in source emplacement (rock type and depth) and regional crustal and upper mantle seismic wave propagation effects. Empirically calibrating regions of interest, as was successfully done for the Soviet test sites in the 1980s, is necessary to provide more precise yield estimations.

Other areas of continuing concern are small and/or over-buried events, which have been found to fail discrimination tests, along with evasive testing of small events (W. Walter, LLNL, briefing to committee, September 4, 2020).³⁵ Waveform correlation (template) methods have been demonstrated to detect very small (magnitude 1.7–3.4) events on and around the North Korean test site (Ford and Walter, 2015; Dodge, 2018). Increasing scatter in observations for earthquakes at very close distances (< 300 km) impacts performance of discriminants for very small events. NNSA is conducting a three-phase Source Physics Experiments (SPE) test bed effort using extensive field recording of chemical explosions at the Nevada National Security Site (NNSS). The explosions are designed to determine effects of source depth, source medium velocity, and pre-existing source region damage. Planned SPE Phase III efforts will include a direct comparison of co-located earthquake and chemical explosion sources (the Rock Valley³⁶ Direct Comparison experiment) designed to eliminate differences in source medium and propagation effects (Walter et al., 2012). Efforts in advanced data analysis to deal with large seismic waveform datasets from multiple events and waveform correlation methods are being developed to detect signals below ambient noise levels.

Another large-scale DNN R&D effort, the Low Yield Nuclear Monitoring test bed and project, is focused on understanding the signals and signatures from low-yield, evasively conducted nuclear test explosions. The project is seeking to develop end-to-end models to predict observables for such tests, and is fusing dynamic and chemical signatures. Identified R&D priorities of the project include (1) analysis of smaller events at shorter distances with higher frequencies; (2) characterization of complex shallow earth structure (local geology is mapped to the extent possible, supplemented by seismic wave speeds and tomography, but detailed near-surface structure is very difficult to determine remotely); (3) distinction between chemical and nuclear explosions; (4) enhancement of correlation methods; (5) data fusion efforts (seismic, acoustic, RN, gravity, electromagnetic, etc.); (6) improvement in yield estimation for events with varying depth (or

___________________

³⁵ The occurrence of very low-yield explosions in new test sites such as in India, Pakistan, and North Korea also raises concern about sub-kiloton tests.

³⁶ Rock Valley is the site of a 1993 earthquake sequence in the NNSS.

height) and emplacement conditions; and (7) identification of characteristics of evasive testing (DNN R&D, communication to committee, September 18, 2020).

3.3.4 Advances in Radionuclide Monitoring

RN monitoring is an important nuclear test explosion MDV technique since detecting fission and activation products confirms whether an explosion detected via waveform technologies (seismic, infrasound, and/or hydroacoustic monitoring) is a nuclear event (CTBTO, 2021a). The CTBTO IMS has planned a worldwide network of particulate (80) and gaseous (40) samplers that will continuously monitor the atmosphere for radionuclides produced via fission (e.g., Xe isotopes) or activation (e.g., ³⁷Ar).³⁷ As shown in Figure 3-4, 95 of these samplers are now operational.

The international RN monitoring community has been working on new developments for both particle and gas systems such that they can sample shorter time periods and enable more accurate RN measurements. Advances in measurement capability have resulted from separate U.S. (Hayes et al., 2018, 2020), French (Cagniant et al., 2018), and Swedish (Fritioff et al., 2017) efforts. After 10 years of development, these next generation Xe and Ar systems are at or near TRL 6, and are in extended CTBTO testing in advance of certification and deployment (H. Miley, PNNL, briefing to committee, September 4, 2020).³⁸

To ensure high sensitivity to nuclear test explosions, characterizing background RN levels is essential. Global Xe levels have risen due to emissions from medical isotope facilities and emerging advanced reactors (Bowyer, 2020). Simulating emissions from known Xe emitters helps guide network design/optimization for realistic background levels. Stack monitors for Xe have been developed and deployed to some of the medical radioisotope production sites with the cooperation of the commercial industry. U.S. researchers are working to develop better atmospheric models to improve stack to station predictions (H. Miley, PNNL, briefing to committee, September 4, 2020; Bowyer, 2020). Xe isotopic ratios change with time and require high-resolution meteorological tracking to help reject sources not related to nuclear explosions.³⁹

Major advances in RN data analysis, undertaken by DNN R&D, aim to enable a new analysis paradigm, focused on identifying “RN events” that may involve multiple sensor stations, rather than focusing on individual detections

___________________

³⁷ Nuclear explosion monitoring focuses on the detection of Xe isotopes (^131mXe, ^133mXe, ¹³³Xe, and ¹³⁵Xe) because they remain in a gaseous state and do not react in the atmosphere. The detection of ³⁷Ar, produced from the activation of ⁴⁰Ca in soil and rock, is complementary to Xe monitoring in the event of an underground nuclear test. See Bowyer et al. (2002) for more information on Xe monitoring and Haas et al. (2010) for more information on Ar monitoring.

³⁸ Next generation Xe systems are expected to be deployed into the IMS as early as 2022 (H. Miley, PNNL, communication to committee, January 13, 2020).

³⁹ This is related to the fate and transport research for environmental monitoring of the nuclear fuel cycle described in Section 3.2.2.

(Eslinger and Schrom, 2016; H. Miley, PNNL, briefing to committee, September 4, 2020). The long-term goal is to fuse particulate and Xe measurements to robustly define events. This new approach is essentially working toward a first principles source model, involving source, containment, and local propagation effects. The approach requires data fusion with high-resolution atmospheric transport modeling, bench experiments, and field experiments. There are currently some bench and small field tests underway, with larger field tests planned in 2021.

Future challenges that must be addressed by the RN monitoring R&D community include characterizing new types of Xe producers (e.g., research reactors, aqueous homogeneous reactors, molten salt reactors), and recognizing explosion-like Xe signatures associated with both reactor start-up and operation of breeder reactors.

3.3.5 Findings and Recommendations⁴⁰

Finding 10. Capabilities for global detection of nuclear explosions have improved since the 2012 National Academies CTBT report. In particular, (1) diverse IMS monitoring networks are approaching the CTBT entry-into-force requirements; (2) extensive analyses of the signals for the underground explosions at the North Korean test site have introduced new source characterization capabilities such as source discrimination with regional waves, full moment tensor analysis of seismic wave radiation, and fusion of seismic and satellite-based ground deformation measurements; and (3) advanced data analytics are being explored in R&D programs for their potential to improve detection capabilities. However, improving detection sensitivity remains a key challenge, as does improving the yield estimate accuracy for uncalibrated test sites and low-yield tests everywhere. In addition, improved transport models for RN back-tracking are needed for high confidence in identification of seismic detections as nuclear explosion sources.

Recommendation 10. NNSA and the Department of Defense should expand support for R&D to improve nuclear explosion detection sensitivity and confidence, as well as yield estimate accuracy. These efforts should include the following:

1. R&D to improve the accuracy of yield estimates from remote measurements for uncalibrated regions of interest and for low-yield explosions at known test sites.
2. R&D to improve detection sensitivity and confidence by developing higher resolution computational transport models (see also Recommendation 8), exploiting all available data sources (including open

___________________

⁴⁰ See also Recommendations 8 and 9 in Section 3.2.3, which are relevant to enhancing MDV capabilities for both the nuclear fuel cycle and nuclear testing.

1. sources⁴¹), and fusing RN monitoring observations with source origin data from seismology or other MDV technologies.

Finding 11. A fully functioning IMS and broader CTBT verification regime is beneficial to U.S. nuclear explosion MDV efforts.

1. CTBTO data are being leveraged, and U.S. support for the CTBTO is being sustained despite non-ratification of the CTBT.
2. International participation in analysis of IMS data is active and there is broad international agreement on the following research needs to improve CTBTO capabilities: understanding atmospheric fate and transport, fusing data streams (e.g., RN and seismic data), characterizing increasing background radiation, filling the data gaps that occur when countries intentionally shut down their sensor network or stop reporting data, and developing an effective on-site inspection capability.⁴²

Recommendation 11. The United States should continue to support CTBTO IMS construction, technology refreshment, and improved IMS capabilities because a fully functioning IMS is beneficial to the United States.

3.4 MDV FOR ARMS CONTROL

Proposals to control and limit nuclear weapons date to the end of WWII. To date, nuclear arms control treaties or agreements have existed only between the United States and the Soviet Union and its successor states, primarily Russia. In the bilateral START I⁴³ and New START⁴⁴ treaties, the verification regimes were negotiated alongside other treaty provisions such as limits, obligations, restrictions, and declarations. In each of these treaties, the treaties limited the number of delivery vehicles as a way to limit deployed nuclear weapons. The number of deployed nuclear weapons were either not limited by the treaty or verified by counting rules,⁴⁵ or techniques were specified to verify something declared not to be a weapon indeed was not (verifying absence). Historically, treaties have included detailed descriptions of the acceptable and approved methods to determine compliance with the treaty. Treaties included multiple types of verification protocols such as data exchanges, notifications, and on-site inspections. Negotiators

___________________

⁴¹ For more information about the use of open-source data for nuclear test MDV, see Section 3.5.

⁴² For more information on CTBT on-site inspections, see Wogman et al. (2011).

⁴³ U.S.-Soviet Strategic Arms Reduction Treaty, signed in 1991, entered into force in 1994, expired in 2009.

⁴⁴ Measures for the Further Reduction and Limitation of Strategic Offensive Arms, signed in 2010, entered into force in 2011.

⁴⁵ Counting rules are a method by which a certain number of nuclear weapons is assigned to a certain type of delivery vehicles. Agreement on counting rules obviates the need to monitor the actual number of weapons.

have included provisions for telemetry exchanges and non-interference with NTM in some treaties as additional confidence building measures.

It is widely accepted in both bilateral and multilateral contexts that NTM, such as national reconnaissance satellites, will be used to supplement cooperative MDV protocols. Today, commercial satellite imagery and other open-source information may complement information provided by NTM for verification of compliance.

The Defense Threat Reduction Agency (DTRA) has the responsibility within the United States to implement and staff bilateral treaty inspections. For the only bilateral treaty constraining nuclear arsenals that remains in force, New START, DTRA maintains the data exchange and notification databases, trains inspectors and provides logistical coordination for the inspector teams, conducts inspections in Russia, and escorts Russian inspectors during their U.S. inspections.

3.4.1 Capability Needs

Because the number of U.S. and Russian deployed strategic nuclear weapons and their delivery vehicles have been constrained by treaties for decades, non-deployed strategic weapons and deployed and non-deployed tactical nuclear weapons are of increased interest in possible future arms control treaties with Russia. It is also expected that any comprehensive treaty or agreement with China would need to include shorter-range systems and associated warheads. Treaties that include weapons in storage or weapons designed for shorter-range delivery systems are anticipated to require new MDV techniques.⁴⁶ At a minimum, such treaties would likely require access to storage areas either directly or remotely, and confirmation of warhead count (either a baseline confirmation⁴⁷ or through routine/challenge inspections). Other needs of such treaties could include tracking movement of warheads or certification of warhead dismantlement. If treaties that decrease the number of nuclear weapons are negotiated, it will be of growing importance to have MDV techniques that are increasingly more robust against attempts to hide warheads by obscuring intrinsic warhead signatures.

In order for tools, equipment, or instruments to be considered to monitor an arms control treaty, the capability needs to be authenticated by the inspector and certified by the host.⁴⁸ Authentication includes assuring that the measurement is as agreed, the equipment functions as designed, and the equipment has no other

___________________

⁴⁶ The 2018 Nuclear Posture Review noted that “the United States remains committed to finding long-term solutions to the technical challenges of verifying nuclear reductions, and therefore will explore new concepts and approaches for this goal” (DoD, 2018, p. 72).

⁴⁷ For additional information, see NTI (2014a).

⁴⁸ The 2014 DSB report assessed that “building greater cooperation and transparency will require trusted technical support systems that do not currently exist” (p. 4). To remedy this, the DSB report recommended that NNSA DNN pursue an international R&D program in automated monitoring and reporting systems supported by information barriers and authentication, and to develop trusted information barriers (DSB, 2014).

function. Certification includes all the same assurances as authentication but also must ensure that the equipment and procedures are safe and that sensitive information is protected. Ideally, MDV techniques would not have undue requirements for the amount of space or time needed for inspections. To date, the MDV techniques selected for deployment have been as simple as possible to ensure treaty obligations are being met and that military-significant cheating by the other side will be identified in a reasonably short period of time. Since inspectors have thus far only needed to verify warhead absence, rather than warhead confirmation, deployed radiation measurement techniques have been passive counters.

3.4.2 Warhead MDV

One of the largest areas for capability development is warhead MDV. The simplest approach to warhead MDV is verifying the absence of a warhead. More complicated techniques are needed to confirm that an item is a warhead while protecting against the transfer of classified or nuclear weapons design information. Finally, because it is anticipated that access to warheads may be limited due to both safety and security concerns, a discussion of technical options for limited access is included.

As discussed in Section 2.3.1, the Warhead Verification Program (WVP) in DNN NPAC’s Office of Nuclear Verification is the focal point of warhead MDV R&D. The WVP is currently funded at $5 million annually. The office has developed a staged approach for MDV needs that follows a warhead lifecycle. The lifecycle for MDV starts with warheads deployed on delivery systems; warheads in storage; warheads being transferred from one facility to another; warhead production, maintenance, and surveillance; and finally warhead dismantlement. With this lifecycle in mind, the WVP developed three approaches:

• Baseline approach that builds from New START and measures the absence of nuclear weapons.
• Additional approach that adds warhead confirmation measurements and dismantlement confirmation, including chain of custody during transport and dismantlement processes.
• Stretch approach that includes making higher fidelity measurements that could support warhead sub-limits and strengthened chain of custody and equipment certification measurements.

This WVP strategy is designed to be mixed-and-matched as needed for negotiated agreements. To date, the Baseline approach has been funded and some Additional approach techniques also receive support. The WVP has not started any significant work toward the Stretch approach.

DNN R&D has historically conducted warhead MDV R&D as well. Over the past several years, DNN R&D worked on the Warhead Measurement Campaign

(WMC), a multi-laboratory collaboration to develop standardized radiation signature data from U.S. nuclear weapons to support treaty verification studies and other efforts (LANL, 2019). However, since the WMC effort ended in FY2018, the DNN R&D Warhead Verification and Monitoring portfolio has only comprised one project. DNN R&D held a meeting in December 2020 to determine future priorities for this portfolio, which could include supporting the WVP stretch goals.

MDV Capabilities for Warhead Absence

In past verification regimes, it has been generally accepted that there was no benefit to declaring a non-weapon item to be a nuclear weapon when the treaty limited the number of weapons. So, only absence measurements were required. To verify absence, typically a neutron or gamma detector is placed near the object of interest to record that no above background radiation emission is detected over a specified period of time. A neutron detector is used to support START and New START. Initially, the New START Treaty used the same detector as START, although the detector was modernized after the treaty entered into force largely to make calibration simpler but otherwise uses the same physics concepts.

In the future, if a passive gamma radiation detector is used, it is likely that a measurement would need to be made to determine the amount of shielding in the container followed by an emissions measurement. In this manner, there could be assurance that shielding has not been deployed to mask the presence of a warhead. The main drawback to gamma passive detection approaches is that ²³⁵U is difficult to detect due to its low energy emission and low emission rates. Therefore passive gamma approaches tend to focus on the detection of ²³⁸U or plutonium (A. Glaser, Princeton, briefing to committee, October 23, 2020).

No information barrier is needed when conducting absence measurements. If the item is not a nuclear weapon but contains nuclear material, then additional measurement capability will likely be needed for the item. If additional measurement capability is needed, techniques for explosives detection or warhead confirmation could be used.

MDV Capabilities for Warhead Confirmation

In order for each warhead to be measured and confirmed, direct access to the weapon or weapon container is required. It is assumed in all R&D work done to-date that weapon design details cannot be revealed (e.g., the ratios of fissile materials).

Two primary techniques have been developed for warhead confirmation, referred to as attribute and template approaches. These approaches use a technique to collect data, algorithmically decide if the weapon is a weapon, and then give a go/no-go signal without sensitive data transfer or revelation.

Attribute approach: The attribute approach uses measurements to satisfy that a previously agreed property indicates a warhead. For example, radioactivity over a set period of time could be measured to confirm the presence and amount of special nuclear material (SNM). If there is more than an agreed-to lower limit of SNM, the technique uses this attribute to confirm the object is a weapon. This approach provides less confidence than approaches that compare more information, but multiple attributes could be measured to provide greater confidence. An example of an attribute system that has been developed and tested is the Trusted Radiation Attribute Demonstration System (Mitchell and Tolk, 2000).

Template approach: Template approaches are more intrusive but may establish more confidence than attribute techniques. They require an exemplar of each weapon type to be measured before monitoring begins and for those data to be kept secret. The measurement information is used to create a weapon template. When the radiation signature of a warhead is complete, the results from the radiation detection equipment are compared to the template. If the results match, then the item is confirmed as a weapon. An example of a template system that has been developed and tested is the Trusted Radiation Identification System (Seager et al., 2001).

There is considerable ongoing research on confirmatory approaches for nuclear weapons MDV, largely aimed at new approaches to information barriers that increase confidence the system cannot be spoofed while protecting sensitive information. The R&D aims to remove sensitive information while providing as much data as possible. Information barriers tend to make authentication and certification difficult; therefore, their sophistication must be balanced against authentication and certification needs.

Early research on information barriers focused solely on electronic encryption; however, this was found to have cyber vulnerabilities. Today, information barrier research is focused on techniques that manipulate the measurement and/or data in ways that make them less vulnerable than electronic encryption. For example, one area of research is physical encryption that aims to disrupt the measurement by inclusion of physical object(s) or manipulation(s) to produce data that are impossible to decrypt.⁴⁹ There is also considerable research on zero-knowledge protocols, which are essentially more sophisticated template approaches. Zero-knowledge protocols, if properly implemented,⁵⁰ cannot reveal sensitive information by design and produce a null set if the object in question

___________________

⁴⁹ The only physical encryption method proposed to data is that of Kemp et al. (2016). At this point, the Kemp method has not been fully operationalized.

⁵⁰ One limitation of physical zero-knowledge protocols currently is that they are intolerant to minor implementation errors. For example, a negative mask must be perfectly aligned with a radiographic image of the warhead to produce a null set and not reveal the edges of the warhead. These protocols also assume that all warheads of a given class are identical, and cannot account for small manufacturing variations between warheads.

matches a trusted template. Several sophisticated zero-knowledge protocols have been proposed recently (Glaser et al., 2014; Marleau and Krentz-Wee, 2017; Hecla and Danagoulian, 2018; Engel and Danagoulian, 2019) but still require significant development and testing before they could be applied in an operational setting. Some of the recent zero-knowledge protocol research use a physical encryption approach along with a template technique.

These new techniques are still in early phases of research, but a need is already arising for a standard test bed that could provide the ability to assess the strengths and weaknesses of different techniques as opposed to relying on the developers to create demonstration tests. There is value in both unclassified and classified test beds that represent real-world settings and real-world objects with, for example, manufacturing variances, temperature fluctuations, or space limitations. Such test beds could eventually stimulate ways to combine the best technical ideas or generate new ideas for specific applications that may arise in arms control negotiations.

Capabilities for Limited Access

The inspected state will likely limit inspector access to any sensitive military sites and nuclear facilities because free access may reveal operational or other national security information. This will be true if verification of non-deployed weapons in storage facilities is included in a future treaty. A limited access regime might be encountered during baseline declaration verifications, routine inspections, or challenge inspections.

If access to the weapon or unique container is provided, serial numbers or unique identifiers could be used to match items with declarations and these same identifiers could be used to track declared items if needed. In a regime like this, it is not necessary to confirm an item declared to be a warhead is a warhead, only to account for it. That is, authenticity of the item is not confirmed, but its provenance is confirmed.

Research to assure that serial numbers have not been tampered with is being conducted by NPAC. Additionally, tags are useful in this type of regime. New tagging concepts have been developed that are inexpensive yet difficult to replicate; an example is the reflective particle tag (Horak et al., 2010). Another approach that is less intrusive and therefore useful in a limited access regime is the “Buddy Tag.” A Buddy Tag is a token created for every declared warhead. The Buddy Tag would be deployed in the same facility as the warhead and can be moved with the warhead. In a challenge inspection, the warhead and the Buddy Tag count would need to be consistent. Buddy Tags have been designed using accelerometers to detect movement (Glaser and Kutt, 2020). Seals, such as those used for safeguards, can be employed if the weapon is placed in a container to assure that the container has not been opened (see Box 3-2 for additional information on tags and seals).

Future research for managing access could include equipment allowing inspections when the inspectors are not present. Such remote inspections would require camera systems and other equipment that would be carried either by the host (inspected) party or by a robot, and perhaps accessible remotely by the inspectors. These systems would then either securely store the information or securely transmit the information to the inspectors. Although there are many technical challenges, such a regime could alleviate some of the concerns regarding inspecting sensitive areas. Managing the access during inspections has traditionally fallen to those that operate the facilities, in either DoD or NNSA. Significant work in red team analysis of managed access provisions will be required.

3.4.3 Warhead Dismantlement MDV

Warhead dismantlement is an especially challenging technical and political issue. Some MDV capability needs for dismantlement include verifying that the fissile material mass entering a dismantlement facility equals the fissile material mass exiting the facility (i.e., a fissile material mass balance for the actual dismantlement process), monitoring the disposition of fissile material from dismantled weapons, monitoring a mass balance and disposition of tritium from dismantled weapons, monitoring the disposition of weapon physics package materials other than fissile materials, and validating that the empty external weapon cases belonged to a nuclear weapon and were not substituted. These and many other issues must be accomplished without disclosing classified weapons information to the other states participating in the dismantlement MDV process.

Within the United States, there have been several warhead dismantlement related projects over the past decades. The WMC that ended in 2018 was a significant effort to lay the groundwork for MDV of warhead dismantlement. Efforts focused on methods and tools to determine what attributes were needed for warhead confirmation and to evaluate different levels of intrusiveness.

International Efforts

Non-nuclear weapons states will likely be interested in nuclear weapon dismantlement to assure that weapons claimed to be dismantled have indeed been dismantled. The participation of non-nuclear weapon states in the dismantlement process introduces a whole range of new challenges that need to be solved. Currently, there are two major international warhead dismantlement efforts underway to facilitate cooperation between non-nuclear weapons states and nuclear weapons states, in addition to advancing technical capabilities. These efforts are discussing MDV protocols for dismantlement and beginning to evaluate technologies to implement the processes. U.S. leadership in these efforts is valuable to provide technical input where the United States has advanced knowledge and to

build confidence that the verification concepts developed take into account U.S. national security and protection of proliferation-sensitive information.

The International Partnership for Nuclear Disarmament Verification (IPNDV), established in 2014, is co-chaired by the DoS Bureau of Arms Control, Verification, and Compliance and the nongovernmental Nuclear Threat Initiative (NTI). The partnership comprises 25+ countries and the European Union and has the mission of developing, testing, and building support for multilateral nuclear disarmament verification approaches that include non-nuclear weapon possessing states as active participants.⁵¹

IPNDV’s goal is to identify challenges associated with nuclear disarmament verification and develop potential procedures and technologies to address those challenges. This work has included defining the 14 distinct steps of the nuclear dismantlement process; from defining and verifying baseline inventory declarations; through the verification of warheads as they are removed from deployed delivery vehicles; through storage and transportation; during dismantlement; and ending after disposition of fissile material, non-nuclear components, and high explosives. To date, IPNDV has included detailed technical analysis of potentially applicable measurement and MDV techniques, table-top and practical exercises, and experimental test beds, in addition to written analysis. IPNDV has produced more than 50 analytical products, including technology data sheets and assessments. The first phase of IPNDV resulted in a key judgment stating, “While tough challenges remain, potentially applicable technologies, information barriers, and inspection procedures provide a path forward that should make possible multilaterally monitored nuclear warhead dismantlement while successfully managing safety, security, non-proliferation, and classification concerns in a future nuclear disarmament agreement” (Dunn, 2017, p. 3). Phase II focused on full descriptions of verifying declarations and how to apply verification across the 14 steps of nuclear weapon dismantlement, in addition to practical exercises and technology experiments. Phase III, which began in 2020, will conduct further hands-on activities, including scenario-based discussions, practical exercises, and technology demonstrations.

The Quad Verification Partnership (the Quad) is a group of nuclear and non-nuclear states (the United States, United Kingdom, Norway, and Sweden) that have been working together since 2015 to explore nuclear dismantlement approaches. The work of the Quad was oriented around preparation, execution, and analysis of the Letterpress live-play exercise in 2017. The Quad agreed in 2020 to continue its collaboration. The verification technologies of interest to the Quad between now and 2025 include

___________________

⁵¹ IPNDV is funded by the partners supporting their own participation and hosting the triannual meetings. In addition, the activities of NTI on behalf of the Secretariat are funded through by NTI, partially through specific government grants by non-U.S. partners. Such donations have come from Canada, the Netherlands, Norway, and Sweden.

___________________

⁵² The INF Treaty authorized the installation and operation of a radiographic imaging system to distinguish banned missiles from non-INF missiles. However, the USSR chose not to install an imaging system at Magna, relying on other inspection methods to confirm that exiting vehicles did not contain prohibited missiles (Harahan, 1993, p. 75).

⁵³ Active interrogation methods introduce additional radiation safety challenges. See NPAC (2017) for more information on mitigating these challenges.

portal monitor that could be used in a nuclear facility for boundary control or as part of an arms control verification regime.⁵⁴ The PMAC system is now an option available for boundary control inside a facility in a hypothetical arms control regime. Research challenges remain for portal monitors. Especially important is the ability to measure the absence or presence of HEU, likely requiring active radiation detection techniques and therefore potentially introducing safety concerns. Also needed are unattended portal monitoring technologies that can detect the absence or presence of explosive materials. Additional research on information barriers and software and electronics verification are needed. New ideas to limit the impact on operations and eliminate administrative control requirements are needed.

There may also be an opportunity to leverage significant investments that the Department of Homeland Security (DHS) has made in radiation portal monitors, albeit for nuclear security rather than arms control applications.⁵⁵ However, portal monitors designed for other applications (e.g., border security) are not ideal for arms control scenarios. For example, extra functionality of commercial off-the-shelf systems can make authentication and certification difficult and these systems could potentially reveal classified information if used in an arms control scenario (A. Swift, Y-12, briefing to committee, October 28, 2020).

3.4.5 Over-the-Horizon Needs

The committee did not conduct data gathering for this interim report on arms control needs beyond monitoring warheads, limited access, dismantlement verification, and perimeter monitoring. Chain-of-custody technologies, for example, have not been addressed; these technologies will be revisited for the final report.

Importantly, new nuclear weapon and related technologies are also being developed by peer competitors. For example, Russia and China are deploying dual-capable missiles. Most notably, the Russian new air-delivered maneuverable ballistic hypersonic missile, the Kinzhal, and the Chinese intermediate range ballistic missile, the DF-26, have both been announced as dual-capable (Putin, 2018;

___________________

⁵⁴ PMAC held two demonstrations, one at Los Alamos National Laboratory (LANL) and one at the Pantex Plant. PMAC was required to be able to be used in unattended or attended operations, be semi-portable, and be reliable. It needed to confirm the presence or absence of 500 g or more of weapons-grade Pu, and the direction of motion of the item. Many anti-tamper or tamper-indicating features were incorporated into the design such as one-layer, dual-sided circuit boards and inability to turn the unit on unless it was fully assembled. To increase host confidence the design was modular, met U.S. and U.K. safety criteria, and contained no software or mechanism to store data. The system was jointly designed.

⁵⁵ The motivation for DHS portal monitors is to prevent smuggling of nuclear or radiological materials into the United States. DHS has 1,400 deployed portal monitors and there have been operational issues associated with nuisance alarms that are still being addressed. For an independent assessment of DHS’s portal monitoring RDT&E, see APS (2013).

Li, 2018). The ability to distinguish between nuclear and non-nuclear armed dual capable missiles is an important topic that will need to be addressed in the future.

In addition, nuclear weapon treaties in the future could well include limits on non-nuclear capabilities such as space weapons or cyber technologies. MDV for these types of capabilities is outside the scope of this report, but it is not too early to think about developing the capabilities for a future treaty or agreement.

3.4.6 Findings and Recommendations

Finding 12. NNSA has maintained a modest portfolio of work in MDV tools for arms control, some of it focused on warhead confirmation measurement completed collaboratively between Defense Programs (DP) and DNN. Recently, the need has increased for MDV technologies for non-strategic and non-deployed warheads in potential new arms control treaties, and significant technical challenges remain.

1. Warhead confirmation techniques that can be practically deployed, authenticated, and certified, especially with trusted information barriers, are not yet mature and would benefit from test beds in order to compare strengths and weaknesses in standard and real-world conditions.
2. Joint U.S.-U.K. R&D has significantly advanced the ability to detect the passage of plutonium through a portal. However, a comprehensive technical solution to portal monitoring is needed that can detect HEU and high explosives in addition to plutonium.
3. The next arms control treaties or agreements may need techniques that rely on warhead identifiers or tags, advanced seals, and possibly new warhead confirmation techniques, especially those that could be used in limited access areas like storage sites. New innovative solutions for such scenarios are still needed.
4. Development of methods to manage access to sensitive facilities and data is needed and must involve the operators of the facilities to be inspected.
5. The proliferation of dual-capable conventional/nuclear delivery systems presents MDV technology challenges that demand attention.

Recommendation 12. DNN’s program for arms control MDV should be a sustained, core element of its program at all TRLs regardless of the current treaty enforcement or future treaty negotiation activity to ensure that the research community is generating and maturing technologies. Collaboration between DP and DNN may be the best way to accomplish some of these efforts.

1. NNSA should establish a U.S. experimental test bed for warhead verification that is accessible to the academic, laboratory, industrial, and international community to safely conduct experiments on real and surrogate materials; help mature technologies; and be subject to red

1. team and white team testing for authentication, certification, managed access, and vulnerability analyses.
2. The NNSA Baseline, Advanced, and Stretch R&D approaches offer a good starting point for investments. However, the Advanced and Stretch research topics will likely take longer to mature. Therefore, the Advanced and Stretch scenarios should be supported in parallel, not in series, with the Baseline work whenever possible.
3. NNSA, in collaboration with DTRA and other interagency partners, should participate in or initiate projects to develop ideas and tools to distinguish conventional and nuclear versions of dual-capable systems for potential future arms control agreements.

Finding 13. Through participation in various international efforts, researchers have had opportunities to develop and test MDV techniques and ideas for weapons dismantlement (including warhead confirmation) without revealing sensitive information with other nuclear weapon states and non-nuclear weapon states.

1. The U.S.-U.K., Quad, and IPNDV programs have been productive venues for international exchange and testing of some MDV techniques. NNSA has not always supported laboratory participation in the IPNDV work at the level required for full participation.
2. There have not been persistent bilateral or multilateral R&D efforts on MDV techniques that involve Russia or China.

Recommendation 13. The United States should remain active in multilateral engagements and seek to increase bilateral engagements to jointly develop technologies for arms control and weapons dismantlement since success ultimately depends on a high level of confidence by both nuclear and nonnuclear states.

1. The United States should re-engage with Russia as soon as possible in joint technical experiments to develop high confidence, authenticatable, and certifiable techniques applicable for future warhead MDV.
2. NNSA should provide support to technology providers to participate in international demonstrations to aid both technology maturation and provide transparency.
3. The United States should apply lessons from the U.K., IPNDV, and Quad partnerships to structure active engagements that include all members of the P5.

3.5 CROSS-CUTTING TECHNOLOGY: LEVERAGING DATA

In recent years, the number of data sources and amount of generated data has grown immensely, including data of relevance to the MDV mission. A great

deal of these data now come from nongovernmental/commercial assets and are available to the public, enabling nongovernmental organizations (NGOs) to make significant contributions to the MDV mission. Open-source assets and data can also be leveraged by government agencies and organizations.

The rise in the amount of both governmental and open-source data has also made data science tools and advanced data analytic methods like AI and ML increasingly important to the MDV mission. Raw data collected by governmental and nongovernmental sources must be analyzed and converted into information of use to decision makers. As datasets grow, this analysis can become too complex or time-consuming to rely solely on human analysts.

In this section, the committee addresses the opportunity for the MDV enterprise to leverage growing amounts of data, taking advantage of nongovernmental/commercial sensing capabilities and data streams, developing data pipelines, and pursuing advanced data analytics R&D. ⁵⁶

As shown in Figure 3-1, open-source assets and data and advanced data analytics play a cross-cutting role across the entire MDV landscape. Not only do they support fuel cycle monitoring, nuclear test explosion monitoring, and arms control monitoring, they can also support MDV in the earliest phases of nuclear weapon acquisition: motivation, capability development, and intent. For example, some current projects in the DNN R&D Data Science portfolio are focused on assessing capability development by detecting and tracking nuclear expertise and research and assessing indicators of intent by distinguishing between benign and illicit use of weapons-usable technologies (DNN R&D, communication to committee, September 18, 2020).

3.5.1 Leveraging Open-Source Assets and Data for MDV

The previous sections of this report have focused primarily on capabilities that the U.S. government or international organizations like the IAEA and CTBTO use to collect MDV data. However, the past several years have also seen a significant increase in nongovernmental/commercial sensing capabilities and resulting data, much of which is available to the public. The government already leverages these capabilities and data to some extent, including for the MDV mission,⁵⁷ but there is still unrealized potential.

Where they can be appropriately verified, open-source data can be another stream of MDV evidence. Open-source data typically come from unauthenticated sources, like commercial or foreign assets, and could therefore be spoofed or tampered with; however, concerns about using unauthenticated data have diminished in recent years as the expanding availability of open data systems provides sufficient

___________________

⁵⁶ See NVCITF (2017) for a prior assessment of how some significant developments in the digital age relate to nonproliferation monitoring by the government and NGOs.

⁵⁷ For example, see Strout (2020) and Hitchens (2020). How the government leverages open-source assets and data for the MDV mission will be addressed in greater detail in the final report.

redundancy and data continuity with which to quasi-authenticate the data. For example, there are straightforward strategies for evaluating signal stability and robustness over time by comparison among large numbers of open-source and NTM recordings for calibration events. Increasing authentication and analysis rates, potentially through the use of advanced data analytics, could increase the utility of open-source data streams. In addition, unauthenticated data can be used to tip and cue more trusted NTM assets for closer examination. In this manner, open-source data are complementary to NTM. In this section, the committee addresses the opportunity for the MDV enterprise to leverage open-source data, focusing on two types of open-source assets with particular relevance to the MDV mission: commercial overhead sensing capabilities and open-source seismic networks.⁵⁸

Commercial Overhead Sensing Capabilities

One area that has changed dramatically with the growth of commercial capabilities is overhead sensing by aircraft or satellite. Today, high-quality overhead sensing data from commercial airborne platforms and satellite constellations are widely available for purchase by the general public. There are 804 earth observation satellites orbiting Earth as of August 2020 (UCS, 2020), 338 (42 percent) of which are commercial, and the pace of adding more and more diverse space-borne sensors continues to accelerate.⁵⁹

Analysts both inside and outside the government use commercial imagery data to support the MDV mission; for example, Box 3-4 summarizes open-source analyses of a North Korean uranium mining and milling site.⁶⁰ The most widely available and by far the most widely and effectively used form of commercial overhead sensing is optical imagery in both visible and non-visible (near-infrared [IR]) wavelengths. However, other sensing technologies are becoming increasingly commercially available and may have application to the MDV mission

___________________

⁵⁸ While the committee focused on commercial overhead sensing capabilities and open-source seismic networks for this report, these examples are not the only types of open-source information with potential relevance to the MDV mission. Other data streams like social media (NTI, 2014b) or trade data (Withorne, 2020; Arterburn et al., 2021) can reveal information of interest to the MDV community. The power of seemingly mundane data streams (e.g., vehicle registration, parking, and frequent flyer databases) was recently demonstrated in Bellingcat’s open-source investigation of the poisoning of Russian opposition leader Alexei Navalny (Bellingcat, 2020). In addition, the growing prevalence of open-source data and social media networks has enabled crowd-sourced analysis (Dietrich, 2014; NTI, 2014b; IAEA, 2020a; M. Hanham and L. Rockwood, Open Nuclear Network, briefing to committee, October 2, 2020).

⁵⁹ In addition to the contemporary data generated from these sensors, there is a significant archive of historical imagery available for analysis. Such historical data have been used to assess former nuclear weapons related activities. For example, see Albright et al. (2020).

⁶⁰ For a review of open-source commercial satellite imagery analysis for the MDV mission, see Albright et al. (2018).

(see Table 3-2).⁶¹ One technology that has garnered recent attention in the MDV community is hyperspectral imaging (see Box 3-5). Understanding the relative availability and respective applications of these technologies is key to forming an integrated strategy for resolving MDV analytic problem sets.

TABLE 3-2 Summary of Types of Overhead Sensing

^a RadarSat, TerraSAR-X, COSMO-SkyMed, ICEYE.

SOURCE: Hanham and Lewis (n.d.).

___________________

⁶¹ RF and radio occultation data are also becoming increasingly available. HawkEye 360 and Kleos are companies offering RF data. Spire offers radio occultation data.

The technologies listed in Table 3-2 are becoming more applicable to the MDV mission due to significant improvements in spatial and temporal resolution. The spatial resolution⁶² of satellite imagery has continued to improve in recent years. In 2000, only one commercial satellite (Ikonos) had a resolution under 2 m, while by 2015, there were about 30 (Pabian, 2015). The highest resolution commercial satellite, DigitalGlobe’s World View-3, now offers imagery at a resolution as low as 31 cm (Maxar, 2021).⁶³

Higher resolution images (Figure 3-5) provide more information for monitoring nuclear activities. For example, 80 cm resolution imagery can be used to detect early stages of new construction (e.g., presence of vehicles, soil movement,

___________________

⁶² Spatial resolution refers to how much physical space is captured in a single pixel.

⁶³ DigitalGlobe was acquired by Maxar Technologies in 2017. DigitalGlobe’s WorldView-4 Satellite, launched in 2016, also provided 31 cm resolution imagery but malfunctioned in 2019 and was determined to be unrecoverable (SpaceNews, 2019). Maxar also offers imagery with synthetic 15 cm resolution (Maxar, 2021).

etc.) or changes at a nuclear test site (e.g., spoil piles, logging trails, vehicle activity). At 30 cm resolution, different vehicle types (sedan versus SUV) can be identified, as well as UF₆ cylinders (Pabian, 2015, p. 16).

Another trend with perhaps even greater significance to the MDV community is the increase in temporal resolution, which refers to how often a location of interest can be imaged. Some commercial providers have now launched constellations of smaller satellites that can collect multiple images of a location of interest at a high temporal frequency. For example, Planet Labs has a constellation of satellites that image the entire planet in one day, every day.⁶⁴ This kind of temporal resolution establishes a baseline to monitor changes and movement, such as patterns of life. This vastly increased tempo of surveillance reduces the ability to mask activity and increases analytic surety.

As discussed in Section 2.3.4, the MDV enterprise needs to stay apprised of rapid technical developments in the commercial sector to identify advancements that could improve MDV capabilities. The enterprise is already engaging with the commercial satellite industry to some extent⁶⁵ and could further leverage these capabilities, along with the data they provide, as addressed further below.

Open-Source Seismic Networks

Significant amounts of open-source data are collected via commercial overhead sensing assets, as described above, but the expansion of public data of use for MDV is not limited to overhead sensing. For example, unauthenticated open-source seismic data from regional or national networks of seismometers operated for earthquake detection are used to supplement data collected by the IMS and USAEDS networks for nuclear explosion detection. Open-source stations may provide closer proximity than possible with the current configuration of NTM assets. Given the standardized nature of global seismic data, these can be readily analyzed by AFTAC.

At present, real-time, on-line data in standardized data formats from more than 3,400 seismic stations around the world are accessible through the data portal of the Incorporated Research Institutions for Seismology (IRIS), which is a consortium of more than 120 U.S. universities funded by NSF. IRIS operates the GSN in collaboration with the USGS. The IRIS data portal also provides

___________________

⁶⁴ Another company, BlackSky Global, is in the process of developing a similar capability. Daily imagery updates of any place on earth are now available via subscription services from imagery suppliers.

⁶⁵ For example, a recent project funded by DNN R&D involved a collaboration between Sandia National Laboratories and BlackSky Geospatial Solutions Inc. to examine how emerging capabilities in the satellite industry can improve remote proliferation detection when combined with subject matter expertise and other analytical capabilities at a U.S. national laboratory (DNN R&D, 2019).

access to data from the Federation of Digital Seismic Networks, and many national networks.

These open-source seismological data streams, totaling well over 7,000 stations around the world, provide seismograms with three-component time-histories of ground motion with precise (GPS/GNSS) timing and known instrument response attributes. Data quality assurance and signal fidelity are monitored continuously to ensure fidelity of the data for research applications. The open-source data have been used to develop models necessary for event locations and seismic magnitude estimators. Additionally, the open-source data are applied by universities to events of interest.

The results are compared with internal NNSA and AFTAC analyses of NTM data for confidence building and assessment of promising new analysis methods. There are also open data sources of meteorological data or other information that can improve atmospheric signal backtracking, and other useful data for identification and characterization applications.

3.5.2 Establishing a Robust Data Pipeline

As demonstrated above, open-source assets and data are of value to the MDV mission space independently of advances in data science. However, emerging data analytical methods can extract additional insight from both open-source and government datasets and strengthen the complementary nature of these different data streams. In order to apply advanced data analytical techniques like AI and ML, data collected from disparate governmental and nongovernmental sources must first be collated and curated via a robust data pipeline.

The committee did not see evidence of such a robust data pipeline, although NNSA has noted the importance of such pipelines. For example, DNN R&D noted that milestones of the Data Science portfolio include “enabling technologies including data management architectures and computing pipelines” (DNN R&D, communication to committee, September 18, 2020). DNN R&D’s Multi-Informatics for Nuclear Operations Scenarios Venture has developed a cloud-based data management infrastructure that is accessible to national laboratory and academic researchers (SEAB, 2020). Another DNN R&D project⁶⁶ is piloting a “cloud-based workspace where program researchers can explore program data and develop and apply advanced data analytics, tap into deep repository of domain knowledge generated by experts at the Laboratories, and easily and efficiently collaborate with multidisciplinary teams across the Laboratory complex” (SEAB, 2020).

The efforts above are a promising step in the right direction, but an effective MDV data pipeline needs to encompass data from the broader MDV enterprise,

___________________

⁶⁶ Ecosystem for Open Science (SEAB, 2020).

potentially including international partners. The data pipeline should have a focal point within the MDV enterprise. NNSA, with its connection to the national laboratories and their high-performance computing capabilities, is the logical choice. Such a data pipeline should be developed under the leadership and coordination of the National Security Council (NSC) to ensure the involvement of all relevant organizations. It is essential that the enterprise work on building this pipeline now in order to support the evolution of data analytics research in the next five years.

The ability to readily share data across the MDV enterprise is an essential component to a data pipeline. However, sharing MDV data across federal departments is challenging due to “classification, sourcing, visibility, legal, and policy barriers” (DNN R&D, communication to committee, October 14, 2020). This is particularly true for information sharing between Title-50 and non-Title-50 organizations. In terms of international engagement, data confidentiality also limits the IAEA’s ability to share data produced in different states (IAEA, 2020a). The 2020 IAEA Emerging Technologies Workshop report noted that “the data sharing restrictions and lack of annotations to train learning-based algorithms is limiting the Department’s ability to leverage AI technologies for safeguards video surveillance analysis” (IAEA, 2020a). Addressing the MDV data pipeline as part of the NSC-led strategic planning process described in Section 2.1.1 could help facilitate data sharing, including with industry, academia, and international partners. Interoperability standards and lexicons can harmonize data definitions across organizations.

3.5.3 Leveraging Advanced Data Analytics for MDV

Once data have been collated and curated, they can be analyzed. Here, advanced data analytics can be employed to provide insight and decision support and allow for a more efficient allocation of resources. The field of advanced data analytics⁶⁷ has seen significant growth in the past decade (MGI, 2016). In particular, advances in AI, ML, data fusion, robotics and autonomy, and computational architectures (e.g., graphics processing units) are driving innovation globally. The government has recognized the need to prioritize R&D in these areas (White House, 2019), and several agencies (and national laboratories using Laboratory Directed R&D [LDRD]) have launched initiatives focused on advanced data analytics.⁶⁸ For example, DOE is planning a focused initiative in AI that seeks to advance the foundation of advanced data analytics, as well as applications to

___________________

⁶⁷ This comes under the rubric of data science and is the integration of statistical and computational sciences.

⁶⁸ For example, NSF, in partnership with the Department of Transportation, DHS, and the Department of Agriculture, launched the National AI Research Institutes program funding seven institutes in 2020 with a second round of competition in 2021 (Gibson, 2020).

mission-specific challenges, including MDV (Service, 2019). In an August 2020 report, the DOE Secretary of Energy Advisory Board AIML task force also noted the need for a DOE-wide AI Capability that is managed at the highest level of the DOE enterprise and encompasses all DOE laboratories, centers, mission areas, and entities (SEAB, 2020).

The NDRD Strategic Plan for FY2020–2024 stresses the need for continued R&D on the application of advanced data analytics to the MDV mission (NSTC, 2019). While advanced data analytics are not silver bullets for the MDV mission and have some significant limitations, addressed below, they can support the MDV mission in a variety of meaningful ways, providing insight and decision support and allowing for a more efficient allocation of resources.

Examples of insight and decision support that advanced data analytics can provide include signature discovery and data fusion. AI and ML can identify new MDV signatures by recognizing patterns that might not be apparent to human analysts. The MDV enterprise is actively exploring this type of application, most prominently through the Advanced Analytics for Proliferation Detection (ADAPD) Venture, DNN R&D’s flagship data science project. Data fusion techniques can integrate data from multiple sensors and correlate multiple signatures to provide higher confidence in event detection. For example, detecting both effluent and seismic signatures would provide greater evidence of a nuclear explosion. Fusing multiple data streams can be especially useful if individual datasets are sparse, as is often the case in the MDV mission space. In addition, data fusion can integrate “traditional” MDV datasets (e.g., data collected from sensors) and “non-traditional” datasets (e.g., patterns-of-life, personal networks, export control logs).

Growing datasets are also increasingly time-consuming for human analysis. Advanced data analytics can enable a more efficient allocation of resources by automating monitoring and performing simple analysis tasks, leaving human analysts free to focus on harder problems. AI algorithms can also be used to allow one sensor to trigger another sensor or direct collections, allowing for persistent monitoring without having to collect continuous data from all sources. DNN R&D is exploring this capability through its Persistent Dynamic Nuclear Activity Monitoring through Intelligent Coordinated Sensing (DyNAMICS) Venture project, which seeks to use knowledge of nuclear processes to “autonomously orchestrate measurements, enabling real-time follow up on detections of proliferation-relevant activity” (DNN R&D, communication to committee, October 14, 2020).

Ongoing MDV Advanced Data Analytics R&D

As shown in Table 3-3, several different types of advanced data analytics applications, including those described above, are areas of current R&D in NNSA

TABLE 3-3 Relevant Applications of Advanced Data Analytics to the MDV Mission Space and Examples of Current NNSA R&D Projects

^a DNN R&D also considers MINOS to be a test bed. See Section 2.2.2 for more information on MDV test beds.

(primarily in DNN R&D).⁶⁹ It is encouraging that DNN R&D has embraced data science R&D and is funding a variety of projects focused on different applications, including multiple multi-laboratory Ventures. The committee was unable to assess each individual project in the DNN R&D Data Science portfolio but was impressed by the projects (ADAPD and MINOS) that were briefed in detail.

The national laboratories are very active in developing advanced data analytics techniques for MDV, with more than 50 LDRD projects currently being supported in this area (see Section 2.3.2 for more information on the LDRD program) (J. Smith, LANL, briefing to committee, September 28, 2020).⁷⁰ The interagency has sought to capture these innovative new ideas. For example, a data analytics workshop hosted by LANL in 2017 brought together interagency

___________________

⁶⁹ While NNSA is not the only agency exploring how data science can support the MDV mission, the committee focused primarily on NNSA’s efforts in this area for this interim report. Within NNSA, data science R&D primarily falls to DNN R&D, which has a data science portfolio that receives approximately $25 million in funding annually. The goal of the portfolio, which focuses on R&D (TRL <5), is to drive the development of the next-generation of AI methods and technologies that can be applied to proliferation detection (DNN R&D, communication to committee, September 18, 2020). DNN R&D noted that there is “significant interaction with interagency partners, particularly, the IC, DoD, and DHS” with respect to this portfolio (ibid.).

⁷⁰ For an example of an LDRD project applying data science to improve proliferation detection, see Thomas (2018).

leaders (DNN R&D, the intelligence community [IC], DoD) and laboratory researchers to discuss how new data analytics methods could be applied to data collections from NNSA experimental test beds (Shirey, 2018).

There are also several NGOs and academic institutions looking specifically at the application of advanced data analytics to open-source datasets (e.g., commercial satellite data, trade data, social media data, academic publications) for nonproliferation (Stewart et al., 2018; Liu et al., 2019; Withorne, 2020; Arterburn et al., 2021).

Of course, these data analytic applications are not unique to the MDV mission. For example, at the 2020 IAEA Emerging Technologies Workshop, a medical university professor discussed how the medical community is using data fusion and deep learning algorithms to provide higher confidence detection of serious medical events (IAEA, 2020a). NNSA and other MDV R&D players could potentially benefit from the best practices from other fields of research to stay on top of emerging cutting-edge capabilities. In addition, there is room to better leverage commercial industry, which is leading innovation in some data science fields, as noted in Section 2.3.4.

Limitations of Advanced Data Analytics to the MDV Mission

While the MDV enterprise is rightfully pursuing advanced data analytics R&D, it is also important to note the limitations of these techniques. Advanced data analytic techniques are very successful at approaching certain types of problems such as image recognition. However, many other MDV problems do not lend themselves as easily to advanced data analytical techniques. For example, advanced data analytics have more limited utility when it comes to physics-based modeling or extrapolation to circumstances that were not included in training datasets. In addition, the black-box nature of some analytical techniques may not be accepted by the MDV enterprise, further constraining applications.

The availability of labeled and unlabeled data is ultimately the limiting factor in the application of advanced analytics to any field, including the MDV mission (JASON, 2017). Large training datasets are essential for advanced data analytics, presenting a challenge for the MDV enterprise, which is data-limited in some respects. Nuclear proliferation has fortunately been a rare event, but this has limited the amount of data that is available to train analytic methods to recognize proliferation signatures. Even under the best condition for data sharing, the nature of MDV activities, test beds, surveillance, and open-source assets means that the data available to train AI applications will be limited in both quantity and quality. Additionally, the data will be sparse, noisy, and unrepresentative on a global scale. MDV research in foundational AI technologies that make effective use of sub-optimal datasets will be needed. These methods will serve to shift the timeline of detection earlier.

3.5.4 Findings and Recommendations

Finding 14. There has been a rapid expansion of commercial remote sensing capabilities over the past decade, both in the United States and abroad. A number of advances support improved MDV:

1. Increased spatial resolution, down to approximately 30 cm, supports more definitive analysis and functional site characterization of existing facilities and the discovery of previously unknown sites.
2. Increased temporal resolution enables monitoring of change over time and increases analytic surety.
3. Increased spectral diversity enables better discrimination of sites, effluents, geology, and other objects of interest.

Finding 15. The amount of open-source data is growing rapidly, along with commercial/nongovernmental processing, exploitation, and dissemination of resulting information. Unauthenticated open-source data have value to MDV efforts, particularly if they are being processed and interpreted by trusted entities such as commercial partners or established academics.

Recommendation 14. Each organization in the MDV enterprise should consider open-source information/data as an important adjunct to NTM that can possibly corroborate or enhance NTM data sources, enable international information sharing at an unclassified level, and/or provide tipping and cueing information for tasking of NTM assets.

1. Operational groups should make sure that they have quick pathways to access useful open-source information when events occur.
2. DNN R&D should consider projects to authenticate open-source information independent of or in collaboration with the open-source information provider.
3. DNN R&D should also continue to explore the potential MDV trade-space between less frequent, higher physical resolution and more frequent, lower physical resolution to see if open-source assets can meaningfully improve monitoring persistence.

Finding 16. Advanced data analytics are rapidly emerging techniques with the potential to facilitate earlier proliferation detection and better decision making.

1. Advanced analytics is of interest to many, if not all, of the organizations that support the MDV mission (DOE/NNSA, DoD, IC, national laboratories, military services, commercial industry, and academia).
2. NNSA/DNN R&D has embraced the importance of advanced data analytics to proliferation detection through its data science portfolio and, in particular, by establishing multi-laboratory projects and Ventures.

Recommendation 15. Advanced analytics R&D efforts within NNSA should be supported with a sustained program and projects beyond the typical three-year lifecycle to allow these efforts to evolve into technology development and deployment efforts that will be of interest to multiple programs and agencies.

Finding 17. Data availability, both labeled and unlabeled, will be the limiting factor in the use of advanced analytics to support the MDV mission. Currently methods are being built from rich U.S. test bed data.

1. To deal with sparse datasets, foundational AI/ML methods need to be developed including the creation and use of synthetic data to train algorithms.
2. Efficient and compliant means to incorporate unclassified information into classified datasets will be essential for maximum data curation and analysis.
3. As these methods move from basic research to practice, they will need to be tested and used in active global scenarios presenting the need for data sharing across organizations and federal departments.

Recommendation 16. The NSC should orchestrate an interagency program to build MDV data pipelines with multi-point data collection and curation, collaborating with international partners where feasible. The committee recommends that the NSC designate NNSA as the lead agency in this effort. This effort should include improving methods for using sparse datasets and physics based modeling, and the ability to merge unclassified and classified data. Establishing a robust data pipeline will take time and, if started now, may result in being able to support the evolution of the data analytics research in five years.

REFERENCES

Arterburn, J., E. D. Dumbacher, and P. O. Stoutland. 2021. Signals in the noise: Preventing nuclear proliferation with machine learning and publicly available information. Washington, DC: Nuclear Threat Initiative (NTI) and Center for Advanced Defense (C4ADS). https://media.nti.org/documents/Signals_in_the_Noise_-_Preventing_Nuclear_Proliferation_with_Machine_Learning__PAI.pdf.

Bellingcat. 2020. Hunting the hunters: How we identified Navalny’s FSB stalkers. https://www.bellingcat.com/resources/2020/12/14/navalny-fsb-methodology/.

Bermudez, J. S., Jr., V. Cha, and B. Lee. 2020. Pyongsan uranium concentrate plant (Nam-chon Chemical Complex). Infrastructure development and status. Washington, DC: Center for Strategic and International Studies. https://csis-website-prod.s3.amazonaws.com/s3fs-public/publication/200826_Bermudez_PyongsanUraniumConcentratePlant_FINAL.pdf.

Bowyer, T. W. 2020. “A review of global radioxenon background research and issues.” Pure and Applied Geophysics. https://link.springer.com/article/10.1007%2Fs00024-020-02440-0.

Bowyer, T. W., C. Schlosser, K. H. Abel, M. Auer, J. C. Hayes, T. R. Heimbigner, J. I. McIntyre, M. E. Panisko, P. L. Reeder, H. Satorius, J. Schulze, and W. Weiss. 2002. “Detection and analysis of xenon isotopes for the comprehensive nuclear-test-ban treaty international monitoring system.” Journal of Environmental Radioactivity 59(2):139-151. https://www.sciencedirect.com/science/article/abs/pii/S0265931X0100042X.

Cagniant, A., S. Topin, G. Le Petit, P. Gross, O. Delaune, T. Philippe, and G. Douysset. 2018. “SPALAX NG: A breakthrough in radioxenon field measurement.” Applied Radiation and Isotope 134:461-465. https://doi.org/10.1016/j.apradiso.2017.06.042.

Cirincione, J. 2007. Bomb scare: The history and future of nuclear weapons. New York: Columbia University Press.

CTBTO (Comprehensive Nuclear-Test-Ban Treaty Organization). 2020. CTBT status of signature and ratification. Vienna, Austria. https://www.ctbto.org/map/ (accessed February 24, 2021).

CTBTO. 2021a. Radionuclide monitoring. Vienna, Austria. https://www.ctbto.org/verification-regime/monitoring-technologies-how-they-work/radionuclide-monitoring/ (accessed February 24, 2021).

CTBTO. 2021b. Seismic monitoring. Vienna, Austria. https://www.ctbto.org/verification-regime/monitoring-technologies-how-they-work/seismic-monitoring/ (accessed February 24, 2021).

Dietrich, C. 2014. Handle with care: Crowd-sourcing and non-proliferation. Paris, France: European Union Institute for Security Studies. http://www.jstor.org/stable/resrep06787.

DNN R&D (Office of Defense Nuclear Nonproliferation Research and Development). 2019. Early proliferation detection FY2019 annual report. Washington, DC: National Nuclear Security Administration.

DoD (Department of Defense). 2018. Nuclear posture review. Arlington, VA. https://dod.defense.gov/News/SpecialReports/2018NuclearPostureReview.aspx.

Dodge, D. A. 2018. “Searching for induced seismicity at Punggye-ri nuclear test site using subspace detectors.” Seismological Research Letters 89(6):2094-2112.

DOE (Department of Energy). n.d. Nuclear fuel cycle. Washington, DC. https://www.energy.gov/ne/fuel-cycle-technologies/uranium-management-and-policy/nuclear-fuel-cycle.

DSB (Defense Science Board). 2014. Task force report: Assessment of nuclear monitoring and verification technologies. Arlington, VA: Department of Defense. https://apps.dtic.mil/dtic/tr/fulltext/u2/a608149.pdf.

Dunn, L. A. 2017. Walkthrough exercise summary. International Partnership for Nuclear Disarmament Verification. Washington, DC: Nuclear Threat Initiative. https://www.ipndv.org/wp-content/uploads/2017/10/IPNDV-Walkthrough-Exercise_FINAL.pdf.

Engel, E. M., and A. Danagoulian. 2019. “A physically cryptographic warhead verification system using neutron induced nuclear resonances.” Nature Communications 10:4433.

Erkman, S., A. Gsponer, J.-P. Hurni, and S. Klement. 2008. The origin of Iraq’s nuclear weapons program: Technical reality and Western hypocrisy. Geneva, Switzerland: Independent Scientific Research Institute. https://arxiv.org/pdf/physics/0512268.pdf.

Eslinger, P. W., and B. T. Schrom. 2016. “Multi-detection events, probability density functions, and reduced location area.” Journal of Radioanalytical and Nuclear Chemistry 307:1599-1605. https://doi.org/10.1007/s10967-015-4339-3.

Ferguson, D. E. 1977. “Simple, quick processing plant.” Memo to F. L. Culler, Intra-Laboratory Correspondence, Oak Ridge National Laboratory, August 30, 1977.

Ford, S. F., and W. R. Walter. 2015. “International Monitoring System correlation detection at the North Korean nuclear test site at Punggye-ri with insights from the source physics experiment.” Seismological Research Letters 86(4):1160-1170.

Fournier, V., and M. Gaspar. 2016. “New IAEA uranium enrichment monitor to verify Iran’s commitments under JCPOA.” IAEA, January 16. Vienna, Austria. https://www.iaea.org/newscenter/news/new-iaea-uranium-enrichment-monitor-verify-iran%E2%80%99s-commitments-under-jcpoa.

Fritioff, T., A. Axelsson, A. Mörtsell, and A. Ringbom. 2017. “SAUNA III new beta detector performance.” CTBT Science and Technology Conference. Vienna, Austria. https://ctnw.ctbto.org/ctnw/abstract/21821.

Gibson, E. K. 2020. “New NSF AI research institutes to push forward the frontiers of artificial intelligence.” National Science Foundation, August 26. Alexandria, VA. https://beta.nsf.gov/science-matters/new-nsf-ai-research-institutes-push-forward-frontiers-artificial-intelligence.

Glaser, A., and M. Kutt. 2020. “Verifying deep reductions in nuclear arsenals: Development and demonstration of a motion-detection subsystem for a buddy tag using non-export controlled accelerometers.” IEEE Sensors Journal 20(13):7414-7421. https://doi.org/10.1109/JSEN.2020.2978540.

Glaser, A., B. Barak, and R. J. Goldston. 2014. “A zero-knowledge protocol for nuclear warhead verification.” Nature 510:497-502. https://doi.org/10.1038/nature13457.

Haas, D. A., J. L. Orrell, T. W. Bowyer, J. I. McIntyre, H. S. Miley, C. E. Aalseth, and J. C. Hayes. 2010. The science case for ³⁷Ar as a monitor for underground nuclear explosions. PNNL-19458. Richland, WA: Pacific Northwest National Laboratory. https://www.pnnl.gov/main/publications/external/technical_reports/PNNL-19458.pdf.

Hanham, M., and J. Lewis. n.d. “Remote sensing analysis for arms control and disarmament verification.” Working Paper of the Nuclear Verification Capabilities Independent Task Force (convened by the Federation of American Scientists). Washington, DC. https://fas.org/wp-content/uploads/media/Remote-Sensing-Analysis-for-Arms-Control-and-Disarmament-Verification.pdf.

Hanham, M., G. Liu, J. Rodgers, M. Best, S. Milne, and O. Lepinard. 2018. Monitoring uranium mining and milling in China and North Korea through remote sensing imagery. CNS Occasional Paper 40. Monterey, CA: James Martin Center for Nonproliferation Studies, Middlebury Institute of International Studies at Monterey. https://www.nonproliferation.org/wp-content/uploads/2018/10/op40-monitoring-uranium-mining-and-milling-in-china-and-north-korea-through-remote-sensing-imagery.pdf.

Harahan, J. P. 1993. On-site inspections under the INF treaty: A history of the On-Site Inspection Agency and treaty implementation, 1988-1991. Treaty History Series. Washington, DC: On-Site Inspection Agency. https://www.dtra.mil/Portals/61/Documents/History/On-Site%20Inspections%20INF%20Treaty-opt.pdf.

Hayes, J., M. Cooper, W. Harper, and M. Panisko. 2018. Xenon International overview. PNNL-27660. Richland, WA: Pacific Northwest National Laboratory. https://doi.org/10.2172/1472068.

Hayes, J. C., C. Aalseth, T. Alexander, H. Back, E. Church, M. Foxe, T. Hossbach, P. Humble, L. Lidey, E. Mace, K. Meinhardt, J. Mendez, J. McIntyre, C. Overman, R. Riemann, A. Seifert, K. Silvers, D. Stephenson, R. Suarez, and G. Whyatt. 2020. “High throughput argon-27 field system.” Pure and Applied Geophysics. https://doi.org/10.1007/s00024-020-02584-z.

Hecla, J. J., and A. Danagoulian. 2018. “Nuclear disarmament verification via resonant phenomena.” Nature Communications 9:1259.

Heinonen, O.. 2020. “Analysis of small particles in support of big decisions.” Belfer Center for Science and International Affairs, June 1. Cambridge, MA: Harvard Kennedy School. https://www.belfercenter.org/publication/analysis-small-particles-support-big-decisions.

Horak, K. E., J. C. Bolles, C. W. Wilson, C. Q. Little, J. A. Romero, R. K. Grubbs, J. C. Bartberger, D. Lee Zamora, A. Gonzales, and P. B. Merkle. 2010. Automated reflective particle tag system for physical authentication. SAND2010-3050C. Albuquerque, NM: Sandia National Laboratories. https://www.osti.gov/servlets/purl/1143513.

IAEA (International Atomic Energy Agency). 1997. Model protocol additional to the agreements(s) between state(s) and the International Atomic Energy Agency for the application of safeguards. INFCIRC/540. Vienna, Austria. https://www.iaea.org/sites/default/files/infcirc540.pdf.

IAEA. 1999. Use of wide area environmental sampling in the detection of undeclared nuclear activities. STR-321. Vienna, Austria.

IAEA. 2011a. Safeguards techniques and equipment: 2011 edition. Vienna, Austria. https://www-pub.iaea.org/MTCD/Publications/PDF/nvs1_web.pdf.

IAEA. 2011b. The nuclear fuel cycle. Vienna, Austria. https://www.iaea.org/sites/default/files/nfc0811.pdf.

IAEA. 2012. “The IAEA safeguards analytical laboratories.” Vienna, Austria. https://www.iaea.org/sites/default/files/safeguardslab.pdf.

IAEA. 2017. Emerging technologies workshop: Trends and implications for safeguards. February 13-16, 2017. Vienna, Austria. https://www.iaea.org/sites/default/files/18/09/emerging-technologies-130217.pdf.

IAEA. 2018. Research and development plan: Enhancing capabilities for nuclear verification. STR-385. Vienna, Austria.

IAEA. 2020a. Application of safeguards in the Democratic People’s Republic of Korea. Report by the director general. GOV/2020/42-GC(64)/18. Vienna, Austria. https://www.iaea.org/sites/default/files/gc/gc64-18.pdf.

IAEA. 2020b. Emerging technologies workshop: Insights and actionable ideas for key safeguards challenges. January 27-29, 2020. Vienna, Austria. https://www.iaea.org/sites/default/files/20/06/emerging-tehnologies-workshop-290120.pdf

JASON. 2009. Rare events. JSR-09-108. McLean, VA. https://fas.org/irp/agency/dod/jason/rare.pdf.

JASON. 2017. Perspectives on research in artificial intelligence and artificial general intelligence relevant to DoD. JSR-16-Task-003. McLean, VA. https://fas.org/irp/agency/dod/jason/ai-dod.pdf.

Kemp, R. S. 2016. “Environmental detection of clandestine nuclear weapon programs.” Annual Review of Earth and Planetary Sciences 44:17-35.

Kemp, R. S., A. Danagoulian, R. R. Macdonald, and J. R. Vavrek. 2016. “Physical cryptographic verification of nuclear warheads.” Proceedings of the National Academy of Sciences 113:8618-8623. https://doi.org/10.1073/pnas.1603916113.

Kovacic, D. N., L. G. Worrall, A. Worrall, G. F. Flanagan, and D. E. Holcomb. 2018. “Safeguards challenges for molten salt reactors.” Institute of Nuclear Materials Management Annual Meeting. Baltimore, Maryland.

LANL (LosAlamos National Laboratory). 2019. STE highlights, November 2019. LosAlamos, NM. https://www.lanl.gov/science-innovation/science-highlights/2019/2019-11.php#:~:text=Warhead%20Measurement%20Campaign%20Team,-LANL%20staff%20received&text=The%20objective%20was%20to%20obtain,nonproliferation%20and%20defense%20program%20needs.

Lewis, J. 2015. “Recent imagery suggests increased uranium production in North Korea, probably for expanding nuclear weapons stockpile and reactor fuel.” 38North, August 12. https://www.38north.org/2015/08/jlewis081215/.

Li, J. 2018. Defense Ministry’s regular press conference on April 26. China Military Online. http://eng.chinamil.com.cn/view/2018-04/27/content_8017126.htm.

Liu, G., J. Rodgers, S. Milne, M. Rowland, B. McIntosh, M. Best, O. Lepinard, and M. Hanham. 2019. Eyes on U: Opportunities, challenges, and limits of remote sensing for monitoring uranium mining and milling. CNS Occasional Paper #44. Monterey, CA: James Martin Center for Nonproliferation Studies, Middlebury Institute of International Studies at Monterey. https://www.nonproliferation.org/wp-content/uploads/2019/01/op44-eyes-on-u.pdf.

Makowsky, P., F. Pabian, and J. Liu. 2020. “Pyongsan uranium mines: Despite typhoons, mining and processing operations continue.” 38North, November 12. https://www.38north.org/2020/11/pyongsan201112/.

Marleau, P., and R. E. Krentz-Wee. 2017. Investigation into practical implementation of a zero knowledge protocol. Technical report. SAND-2017-1649, 1357490. Albuquerque, NM: Sandia National Laboratories. https://doi.org/10.2172/1367490.

Maxar. 2021. Satellite imagery. https://www.maxar.com/imagery-leadership.

McLaren, R. 2016. The Independent Person Report. https://www.wada-ama.org/sites/default/files/resources/files/20160718_ip_report_newfinal.pdf.

MGI (McKinsey Global Institute). 2016. The age of analytics: Competing in a data-driven world. https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/the-age-of-analytics-competing-in-a-data-driven-world.

Mitchell, D., and K. Tolk. 2000. Trusted Radiation Attribute Demonstration System. SAND20001481C. Albuquerque, NM: Sandia National Laboratories. https://inis.iaea.org/collection/NCLCollectionStore/_Public/32/016/32016771.pdf.

NASEM (National Academies of Sciences, Engineering, and Medicine). 2005. Monitoring nuclear weapons and nuclear-explosive materials: An assessment of methods and capabilities. Washington, DC: The National Academies Press.

NASEM. 2012. The Comprehensive Nuclear Test Ban Treaty: Technical issues for the United States. Washington, DC: The National Academies Press.

Nikitin, M. B., A. Andrews, and M. Holt. 2012. Managing the nuclear fuel cycle: Policy implications of expanding global access to nuclear power. RL34234. Washington, DC: Congressional Research Service. https://fas.org/sgp/crs/nuke/RL34234.pdf.

NPAC (Office of Nonproliferation and Arms Control). 2017. Joint U.S.-U.K. report on technical cooperation for arms control. Washington, DC: National Nuclear Security Administration.

NSTC (National Science and Technology Council). 2019. Nuclear defense research and development strategic plan for fiscal years 2020–2024. Washington, DC.

NTI (Nuclear Threat Initiative). 2014a. Innovating verification: New tools and new actors to reduce nuclear risks; Verifying baseline declarations of nuclear warheads and materials. Washington, DC. https://media.nti.org/pdfs/WG1_Verifying_Baseline_Declarations_FINAL.pdf.

NTI. 2014b. Innovating verification: New tools and new actors to reduce nuclear risks; Redefining societal verification. Washington, DC. https://media.nti.org/pdfs/WG2_Redefining_Societal_Verification_FINAL.pdf.

NTI. 2015. “Chain of custody, tags, seals and tamper-indicating enclosures.” Washington, DC. https://www.nti.org/analysis/articles/tags-seals/.

NVCITF (Nuclear Verification Capabilities Independent Task Force). 2017. Monitoring and verification in the digital age: Seven recommendations for improving the process. Washington, DC: Federation of American Scientists. https://fas.org/wp-content/uploads/media/Nuclear-Monitoring-and-Verification-in-the-Digital-Age.pdf.

Pabian, F. 2015. Commercial satellite imagery as an evolving open-source verification technology: Emerging trends and their impact for nuclear nonproliferation analysis. JRC technical reports. Luxembourg. https://ec.europa.eu/jrc/en/publication/commercial-satellite-imagery-evolvingopen-source-verification-technology-emerging-trends-and-their.

Pabian, F. V., P. Makowsky, and I. Buck. 2018. “North Korea’s uranium mining and milling operations continue at Pyongsan.” 38North, November 2. https://www.38north.org/2018/11/pyongsan110218/.

Putin, V.. 2018. “Presidential address to the Federal Assembly.” March 1. Moscow. http://en.kremlin.ru/events/president/news/56957.

Robinson, C. P. 2016. “The joint verification experiment and the nuclear testing talks.” In Doomed to cooperate, edited by S. S. Hecker, 85-98. Los Alamos, NM: Bathtub Row Press.

Rockwood, L., N. Mayhew, A. Lazarev, and M. Pfneisl. 2018. IAEA safeguards: Staying ahead of the game. 2019:14. Swedish Radiation Safety Authority. https://vcdnp.org/wp-content/uploads/2019/09/201914-iaea-safeguards-staying-ahead-of-the-game.pdf.

Sagan, S. D. 2011. “The causes of nuclear weapons proliferation.” Annual Review of Political Science 14:225-244.

Schmerler, D. 2020a. “Roof repair work spotted at the Pyongsan uranium mill.” Arms Control Wonk blog, April 17. https://www.armscontrolwonk.com/archive/1209263/atomic-jjimjilbang/.

Schmerler, D. 2020b. “A satellite imagery review of the Pyongsan uranium mill.” Arms Control Wonk blog, July 1. https://www.armscontrolwonk.com/archive/1209625/pyongsan-uranium-mill/.

Schoeppner, M. 2018. Remote detection of undeclared reprocessing. International Panel on Fissile Materials.

SEAB (Secretary of Energy Advisory Board). 2020. Final report of the SEAB to Secretary of Energy Dan Brouillette regarding the Department of Energy and artificial intelligence and machine learning. Washington, DC. https://www.energy.gov/sites/prod/files/2020/11/f80/SEAB_AIML_Report.pdf.

Seager, K. D. et al. 2001. “Trusted radiation identification system.” 42nd Annual Institute of Nuclear Materials Management Annual Meeting, July 15-19, 2001, Indian Wells, CA.

Service, R. F. 2019. “Department of Energy plans major AI push to speed scientific discoveries.” Science, October 30. https://www.sciencemag.org/news/2019/10/department-energy-plansmajor-ai-push-speed-scientific-discoveries.

Shirey, R. 2018. Nuclear Nonproliferation and Security Program Office: Highlights 2017. Los Alamos, NM: Los Alamos National Laboratory. https://permalink.lanl.gov/object/tr?what=info:lanlrepo/lareport/LA-UR-18-22521.

SpaceNews. 2019. “DigitalGlobe loses WorldView-4 satellite to gyro failure.” January 7. https://spacenews.com/digitalglobe-loses-worldview-4-satellite-to-gyro-failure/.

Stewart, I. J., A. Lee, and A. ElGebaly. 2018. “Automated processing of open source information for nonproliferation purposes.” Journal of Nuclear Materials Management 46(3):21-36. https://resources.inmm.org/system/files/jnmm/vol_46/V-46_3.pdf.

Strout, N. 2020. “Inside the intelligence community’s new plan for commercial imagery.” C4IS-RNET, July 17. https://www.c4isrnet.com/intel-geoint/2020/07/17/how-the-intelligencecommunity-is-approaching-commercial-imagery/.

Szymanski, J. J., and P. G. Weber. 2005. “Multispectral thermal imager: Mission and applications overview.” IEEE Transactions on Geoscience and Remote Sensing 43(9):1943-1949.

Thomas, J. 2018. “Researchers developing deep learning system to advance nuclear nonproliferation analysis.” Lawrence Livermore National Laboratory, August 21. Livermore, CA. https://www.llnl.gov/news/researchers-developing-deep-learning-system-advance-nuclearnonproliferation-analysis.

UCS (Union of Concerned Scientists). 2020. Satellite database. Updated August 1, 2020. Cambridge, MA. https://www.ucsusa.org/resources/satellite-database.

Walter, W. R., and L. Wen (eds.) 2018. “Focus section on North Korea’s September 2017 nuclear test and its aftermath.” Seismological Research Letters 89: 6. https://pubs.geoscienceworld.org/srl/issue/89/6.

Walter, W. R., M. L. Pyle, S. R. Ford, S. C. Myers, K. D. Smith, C. Snelson, and V. Chipman. 2012. Rock Valley direct earthquake-explosion comparison experiment (RV-DC): Initial feasibility study. LLNL-TR585672. Livermore, CA: Lawrence Livermore National Laboratory.

White House. 2019. Executive order on maintaining American leadership in artificial intelligence. Washington, DC. https://www.whitehouse.gov/presidential-actions/executive-ordermaintaining-american-leadership-artificial-intelligence/.

Withorne, J. 2020. Machine learning applications in nonproliferation: Assessing algorithmic tools for strengthening strategic trade controls. Monterey, CA: James Martin Center for Nonproliferation Studies, Middlebury Institute for International Studies at Monterey. https://nonproliferation.org/wp-content/uploads/2020/08/CNS_Nonpro_Notes_082020.pdf.

WLN (World Nuclear News). 2018. “Silex Systems out of GLE restructure.” June 13. https://www.world-nuclear-news.org/UF-Silex-Systems-out-of-GLE-restructure-1306187.html.

Wogman, N. A., R. F. Payne, J. I. Friese, T. W. Bowyer, Y. Onishi, M. S. Wigmosta, B. D. Milbrath, C. E. Seifert, H. S. Miley, R. C. Hanlen, and J. C. Hayes. 2011. A discussion of procedures and equipment for the comprehensive test ban treaty on-site inspection environmental sampling and analysis. PNNL-20151. Richland, WA: Pacific Northwest National Laboratory. https://www.pnnl.gov/main/publications/external/technical_reports/PNNL-20151.pdf.

Worrall, L. G., A. Worrall, G. F. Flanagan, S. Croft, A. M. Krichinsky, C. A. Pickett, R. D. McElroy Jr., S. L. Cleveland, D. N. Kovacic, J. M. Whitaker, and J. L. White-Horton. 2016. “Safeguards considerations for thorium fuel cycles.” Nuclear Technology 194:281-293.

Yocky, D. A., R. D. West, R. Riley, T. M. Calloway, D. E. Wahl, L. Garley, and S. A. Bolin. 2018. Detecting surface change created by an underground chemical explosion using fully polarimetric VideoSAR. Albuquerque, NM: Sandia National Laboratories. SAND2018-12625R. https://www.osti.gov/servlets/purl/1481501.

This page intentionally left blank.