Estimating North Korea's Nuclear Material Inventory

What nuclear fuel cycle modelling and open sources tell us about North Korea’s nuclear ambitions

By Grant Christopher, Hailey Wingo, Dave Schmerler and Hugh Chalmers

The North Korean nuclear arsenal continues to be a major challenge to the international community. Since its first nuclear weapon test in 2006, the Democratic People’s Republic of Korea (DPRK) has expanded its nuclear arsenal, despite repeated demands from the United Nations Security Council to abandon their nuclear weapon programme in a complete, verifiable and irreversible manner. The prospects for this are bleak. But if an opportunity presents itself to restrain or roll back the DPRK’s nuclear weapon programme, the international community should be prepared with a holistic view of that programme, what weapon-usable materials it has produced, and how it might evolve. This story shows how fuel cycle modelling and open source information can be combined to explore different scenarios for the DPRK’s current inventory of weapon-usable material, and how that inventory might change.

This story summarises the findings of two more detailed stories, regarding the DPRK's production of highly-enriched uranium ( here ) and its production of separated plutonium ( here ).

The international community has always had limited knowledge of the DPRK nuclear programme. The DPRK implemented comprehensive IAEA safeguards slowly and reluctantly, and ultimately withdrew from those arrangements when proliferation concerns were raised. The window onto the DPRK’s nuclear programme opened significantly first with the Agreed Framework and then with the Six Party Talks – but has since shrunk to almost nothing.

The IAEA and the United States were able to conduct monitoring and inspections while the window was open, but much of this is confidential or is classified. Open sources of information (such as satellite imagery and DPRK media sources) now play a central role in monitoring the DPRK weapon programme and its likely evolution.

We demonstrate here how open source information and a numerical model of the DPRK nuclear fuel cycle (built through the UK National Nuclear Laboratory’s Orion Software) can be combined to explore different fuel cycle development scenarios, and their impact on the DPRK’s inventory of weapons-usable nuclear material.

How do you model a nuclear fuel cycle?

The two primary materials required for fission in a nuclear weapon are plutonium and highly-enriched uranium (HEU). Uranium occurs naturally, but in a form that is not directly usable for nuclear weapons. To be weapons-usable, uranium must be highly-enriched to decrease the proportion of its most abundant isotope (uranium-238) and increase the proportion of its fissile isotopes (primarily uranium-235). Plutonium does not occur naturally, and must be produced by irradiating uranium with neutrons until it decays into plutonium. The nuclear infrastructure required to extract uranium, convert it, fabricate it into useful forms, enrich it, irradiate it, and then process spent fuel is known as a nuclear fuel cycle - and is illustrated in Figure 1 below. This nuclear fuel cycle also provides the DPRK with the opportunity to produce weapons-usable HEU and separate weapons-usable plutonium from spent reactor fuel.

Figure 1: An illustration of a generic nuclear fuel cycle, in which facilities for processing and irradiating uranium are also used to generate weapons-usable HEU and Plutonium

We have built a model of North Korea’s nuclear fuel cycle using ORION: a systems dynamics nuclear fuel cycle simulator developed by the UK National Nuclear Laboratory (NNL). It can simulate all the facilities in a nuclear fuel cycle that process nuclear material, with a graphical user interface to lay out and connect fuel cycle facilities on a canvas (as shown in Figure 2 below). This allows us to reflect known facilities in the DPRK's nuclear programme, and incorporate other potential unknown facilities. When provided with operational parameters – such as availability of feed materials, the operational characteristics of facilities, and operating timeframes – it can simulate the flow of nuclear material through North Korea’s fuel cycle and generate estimates of the weapons-usable material that would be produced. The software can simulate and track all isotopes of uranium and plutonium, but the estimated stockpiles we present here are simplified to the total element weight of all of these isotopes.

Figure 2: A screenshot from the NNL ORION software, illustrating our model of the DPRK nuclear fuel cycle

Without access to the full operational parameters of the DPRK’s nuclear fuel cycle, we have to estimate operating parameters and accommodate uncertainties in these estimates. We use a Monte Carlo method to do this: we assign parameters (such as thermal power of a reactor) a probability distribution across a certain range, using open source information to constrain the likely parameter values within that range. For example, a ‘top hat’ distribution is used to define the start and end dates of reactor operating windows, with each start/end date in that range being equally likely. Depending on the scenario we model, up to 70 parameters are fed into the model. By supplying ORION with a large range of operational parameters, we can generate estimates of the DPRK’s stockpiles of plutonium and HEU that reflect our uncertainties in those operational parameters (see Figure 3 below).

This model uses Orion software, which has been developed by the National Nuclear Laboratory Ltd. The results of the model are driven by assumptions and data selected by the authors and therefore should only be viewed as an indication the DPRK fuel cycle.

Figure 3: Examples of estimated separated plutonium stockpile growth, and HEU stockpile growth for different enrichment capacities

Why use a fuel cycle model?

This model provides a platform from which we can build an evolving and responsive understanding of the DPRK fuel cycle. If new information becomes available regarding known facilities (such as a new operational cycle of the 5 MWe Magnox-type reactor at Yongbyon) or previously unknown facilities (such as unknown uranium enrichment plants or the expansion of the Uranium Enrichment Plant at Yongbyon), that information can be incorporated into the model’s parameters to update our estimates.

By treating the fuel cycle as a system, we can evaluate the trade-offs between different parts of the fuel cycle. If uranium is enriched for the Experimental Light Water Reactor (ELWR) for instance, enrichment capacity is taken up that would otherwise be used to make HEU for weapons. Similarly, we can examine to what degree the North Korean programme is constrained by uranium for the HEU and plutonium routes.

The model also provides a tool to test hypotheses or assumptions about how the DPRK’s fuel cycle may have evolved, or could evolve in the future. By constructing multiple scenarios that are consistent with open sources and modelling them, the implications and differences of these scenarios can be explored. Similarly, the model can consider an assumed change in operational parameters (such as possible restrictions on reactor operation) and explore how that would impact the production of weapons-usable nuclear material.

The model can also be used to prioritise ongoing research into the DPRK’s nuclear fuel cycle. A sensitivity analysis of the model’s estimations can highlight those operational parameters that have the most significant impact on the DPRK’s possible stockpiles and the largest degree of uncertainty and therefore may be a priority to determine through future inspections.

What do we know about the DPRK's fuel cycle?

The primary features of the DPRK’s nuclear fuel cycle are well-established, thanks to the access provided to IAEA inspectors and international experts through the Agreed Framework and to the end of the Six Party Talks era. The DPRK is known to operate the vast majority of facilities that make up the nuclear fuel cycle described above: it has facilities to mine, mill, convert, and fabricate uranium into fuel for a 5 MWe Magnox-type reactor. It has the facility to reprocess the spent fuel from that reactor to separate weapons-usable plutonium. It also has the facility to enrich uranium into HEU for weapons purposes. It is also known to be pursuing an experimental light water reactor. Follow the map below to understand the sites that have been identified as part of North Korea’s fuel cycle.

The interactive Map Tour below gives you a birds-eye view of what is commonly known of the DPRK's nuclear fuel cycle.

01 / 11

Mining: Pyongsan Uranium Mines

In 1992, the DPRK declared two mines to the IAEA: Sunchon/Wolbisan and Pyongsan. An IAEA visit prior to the Agreed Framework declared that the material mined was anthracite coal, with nickel and vanadium by-products. The type of ore mined at Pyongsan, and its uranium grade, has been the result of considerable debate, with no conclusive answers.

Milling: Pyongsan Uranium Mill

In a 1992 site visit by the IAEA, Pyongsan is described as a nickel, vanadium, and uranium mill with anthracite coal as the raw input material for the mill. David Schmerler provided a new explanation of the process flow at Pyongsan mill consistent with the facility processing anthracite coal. In this analysis, the mill is a uranium mill with vanadium and nickel by-products where uranium is extracted from coal-ash after production in a rotating kiln. One of the insights of this analysis is that the area covered by the Pyongsan mill site is large relative to other uranium mills with similar production estimates. Understanding the site as a multi-material facility we estimate annual production as 10-30 tonnes of uranium ore concentrate per year from 1990-1993 as a commissioning phase, and 50-120 tonnes per year from 1994 to the present day.

Milling: Pakchon Uranium Mill

Pakchon is a former phosphate/rare earth mill that was converted to a uranium mill. It operated from 1982-1992 as a pilot plant for yellowcake production. Estimated cumulative production at the mill from open sources is 210 tonnes of uranium ore concentrate. It should be noted that this was the only known operational mill to fuel the 5 MWe reactor at Yongbyon in 1986.

Conversion

All known conversion facilities are located at the Yongbyon nuclear complex. However, North Korea may operate additional conversion plants at unknown locations. For the fuel cycle model, the availability of conversion facilities does not provide a constraint on the model but can provide contextual information for other stages of the fuel cycle.

Jamie Withorne (at the time a CNS Research Assistant) has written a series of posts on North Korea's uranium conversion programme. Please click on the link below to read more.

DPRK Uranium Conversion

Uranium Enrichment Facility

The Uranium Enrichment Facility is housed in what was once the Fuel Fabrication Facility for the 5 MWe Magnox-type reactor at Yongbyon. In 2010, Siegfried Hecker visited with a delegation from Stanford University, and discovered gas centrifuges which can be used to enrich uranium to HEU.

Potential Magnox Fuel Fabrication Plant

Previously, the Magnox Fuel Fabrication Plant was located in the same area as the uranium conversion facilities. The picture above was taken of the old Fuel Fabrication Plant when Hans Blix, then IAEA Director General, visited in 1992.

That building, in which metallic uranium roads were assembled into Magnox-clad fuel elements, is now the Uranium Enrichment Facility, and there is speculation that the fuel cladding process was relocated to a location north of the 5 MWe reactor.

5 MWe Magnox-type Reactor

The 5 MWe reactor is a Magnox-type reactor located at Yongbyon and has been operating intermittently since 1986. Other than the IRT, it is the only reactor North Korea has operated capable of producing plutonium. Typically a full fuel load of 50 tonnes of natural uranium is assumed to be in the reactor for 2-3 years, depending on the observed operational cycle.

Radiochemical Laboratory

The radiochemical laboratory (RCL) at Yongbyon is the main facility at Yongbyon, with the first process line tested in 1990 and the second completed in 1994. This plant uses a PUREX process and has an estimated capacity of 110 tonnes of uranium (or 220-250 tonnes depending on the source). This is far more than the required capacity for the 5 MWe Magnox-type reactor of 50 tonnes of uranium per year.

Potential Fuel Fabrication Plant for ELWR

It is possible that the fuel fabrication for both the 5 MWe reactor and the ELWR are co-located in the same facility (location 6 on this map). However, an ISIS report identified this site as another potential facility for ELWR fuel fabrication, citing an unnamed government source that detected the delivery of materials to this site which are used to produce sintered LEU pellets. These pellets can then be inserted into fuel cladding to produce LWR fuel.

Experimental Light Water Reactor

The ELWR, up until the end of 2022 was not operational, but has recently been a source of renewed analysis about its status of operation. Fuel was thought to be fabricated for this reactor in the last decade and it is thought that the facility requires four tonnes of uranium fuel, enriched to 3.5%.

IRT-2000 Research Reactor

The IRT was built in the 1960s and was supplied with Soviet-manufactured fuel until recently; it now uses a mixed core load including 80% HEU. The exact uses of the reactor are not known, nor is it known whether the DPRK has manufactured HEU fuel for the reactor since a 1987 upgrade to 8 MWth.

What we do not know about the DPRK's fuel cycle

Unknown or unidentified facilities

The DPRK has declared fuel cycle facilities that have yet to be identified in open sources, such as the uranium mine at Sunchon/Wolbisan. There are also indications that the DPRK operates a second gas centrifuge enrichment plant, potentially at Kangson. While a candidate facility has been identified in that area (see the image to the right) it is not confirmed whether it carries out enrichment or is merely associated with enrichment. There may be many other facilities associated with other sections of the fuel cycle that have yet to be revealed, and these could significantly expand the capacity and nature of the DPRK’s fuel cycle. There is no recipe that can be followed for identifying such facilities: it is an artisanal, labour-intensive process that requires cross checking multiple sources, using experienced analysts and requires access to satellite imagery processing tools and databases. But for all the challenges in finding unknown facilities, there have been remarkable results from NGOs using open sources, particularly with remote sensing, to uncover nuclear-related facilities. Our model simulates the existence of an unknown enrichment facility in some scenarios, and demonstrates how identifying such a facility changes the picture of the DPRK’s programme.

Operational history

Even while international inspectors had access to the known DPRK facilities, it was not always possible to confirm when those facilities were operating and for how long. The activation and deactivation of reactors signals when irradiated fuel could be removed for cooling and reprocessing to separate out weapons-usable plutonium. The time in which uranium mines and mills became operational can have a significant impact on the stockpiles of uranium source material that the DPRK’s fuel cycle can call upon. Satellite imagery can reveal some indicators of operational status, such as heat radiating from active facilities, tertiary cooling water being discharged from active reactors, or the presence of cooling units to manage operating temperatures. Not all operational activities present signatures that are detectable through commercially-available remote sensing, and where signatures are detected it is hard to extrapolate those over a period of time. We have collected and share here evidence from open sources to inform estimates of the operational history of known facilities, and demonstrate how different operating assumptions change the picture of the DPRK’s programme.

Facility design and sophistication

The operating capacity and efficiency of the DPRK’s nuclear facilities is very likely to evolve over time, driven upwards by operational experience and design improvements but also degraded by ageing facilities and incidents that could disrupt operations. Operational losses inevitably arise in any bulk processing of nuclear material, and the scale of those losses depends on the equipment being used and the manner in which it is used. These factors are very challenging to estimate from the outside, but the DPRK has been proud to show off some of its improvements – such as the incorporation of pulse columns into reprocessing facilities – to visiting international experts. Visiting international experts were able to estimate the rough technology type (and therefore effectiveness) of the DPRK’s gas centrifuges when they were presented to them in 2010, but there is little open source information about how that technology may have evolved since then. We have seen the evolution of similar centrifuge technology in Iran and include a scenario with a conservative estimate of centrifuge improvement to demonstrate how changes in that technology change the picture of the DPRK’s programme.

Strategic fuel cycle choices and trade-offs

The DPRK’s nuclear fuel cycle does not necessarily serve the sole purpose of accumulating weapons-usable nuclear material. DPRK officials claimed during a visit of international experts in 2010 that the Experimental Light Water Reactor (ELWR) under construction at Yongbyon is primarily for the production of electricity. This implies that, once operational, the refuelling frequency of that reactor may be lower than if it were primarily intended for plutonium production (although the model assumes that the ELWR is not yet operational). Similarly, in the same visit DPRK officials claimed that the IRT research reactor is very important for domestic medical radioisotope production, but is running low on fuel. This implies that the DPRK may divert some of their uranium, enrichment and fuel fabrication resources towards providing a domestic source of fuel for that reactor – at the expense of increasing their stockpile of HEU. The DPRK can apply its uranium and fuel cycle resources to achieve a number of objectives, and the choices it makes entail trade-offs about where those resources are directed or not directed, and how much weapons-usable material it can produce as a result. We have simulated the consequences for the DPRK’s stockpile of weapons-usable material for a variety of choices, including the fuelling of the ELWR, the fuelling of the IRT reactor, and the adjustment of fuelling for the 5 MWe reactor to demonstrate how these choices change the picture of the DPRK’s programme. Figure 4 to the right demonstrates how the DPRK's uranium output supplies (on the right) are allocated to different fuel cycle activities, and the end products resulting from those activities.

Modelled scenarios

The DPRK's stockpile of weapons-usable plutonium (Pu)

We assume that the 5 MWe reactor at Yongbyon is the only significant source of plutonium in the DPRK, and it is the only source considered in our model. We assume that plutonium in the fuel of the IRT-DPRK research reactor is not of sufficient isotopic quality for the DPRK's weapon purposes, and that the DPRK has not adjusted its reprocessing plants to handle this fuel. We also assume that the DPRK is not utilising the experimental channels in the research reactor to irradiate uranium targets for plutonium production.

We assume that the Experimental Light Water Reactor (ELWR) is not operational, and that the DPRK would have to carry out a number of time-consuming (but not insurmountable) steps to separate plutonium from its fuel once it becomes operational.

These assumptions imply that in the medium-term, the DPRK's stockpile of weapons-usable plutonium is dependent on the on-going fuelling and operation of the 5 MWe reactor, and its reprocessing plant. The growth of this inventory is illustrated in Figure 5 to the left.

Click here for more information

The DPRK's stockpile of highly-enriched uranium (HEU)

Our modelling of the DPRK's HEU stockpiles is dependent on a much broader array of scenarios than its plutonium stockpile. There are significant uncertainties about the number of centrifuges that the DPRK can call upon, how long those centrifuges have been operating, and how effective they are. It is also unclear what demands - other than HEU production - might be placed on this enrichment capacity. We have accommodated these uncertainties by simulating scenarios that consider:

Increasing demands on uranium resources and enrichment capacity from fuelling the ELWR, the IRT-DPRK reactor, and modifying the fuel for the 5 MWe reactor for tritium production.
For each of these demand scenarios, we simulate sub-scenarios in which the DPRK can also call on additional centrifuges at an unknown facility, at varying times.
For each of these sub-scenarios we simulate further sub-sets in which the DPRK can call upon more advanced and effective centrifuges than the first-generation technology revealed to international experts.

We have simulated 48 HEU production scenarios in total, and the effect of these uncertainties are illustrated in the interactive Figure 6 to the left.

Click here for more information

Forecasting work

The stockpile estimates above extend to the end of 2023 and build on what we know about the history and current state of the DPRK’s nuclear fuel cycle. However, it is also important to consider what nuclear capabilities the DPRK could be pursuing, and what this implies for its fuel cycle and stockpiles of nuclear material. We can simulate the future fuel cycle, but this requires a different type of input in the absence of historical open-source information. To collect the information needed, VERTIC and CNS, along with the One Earth Future Foundation’s Open Nuclear Network (ONN) hosted a workshop in February 2024. The workshop convened experts with a broad array of relevant technical and policy expertise.

During the workshop, participants provided a baseline for forecasting by outlining the key objectives of the DPRK’s nuclear programme: defence, technical development, bargaining leverage, and prestige; as well as potential drivers of change for the nuclear programme: nuclear accidents, famine or pandemic, economics, recognition as a nuclear state, change in leadership or regime collapse, and change in alliances.

Forecasting work on the impact of these drivers was facilitated by the Swift Centre, who used the Delphi method and provided a platform to track participants’ anonymised estimates and virtual discussion. Ongoing discussion took place simultaneously, as participants interrogated their assumptions and debated the impact of potential future drivers on various choices and limitations for the DPRK’s nuclear programme. To translate qualitative discussion into measurable results, we used a baseline forecasting question of how many nuclear warheads the DPRK would have in its arsenal by 2029 and how this number might change taking into account the discussed drivers of change. This controlled methodology enabled discussants to refine their initial estimates based on new information or perspective provided by other experts in the room, but not so frequently as to encourage conformity bias.

Projections of the DPRK's nuclear material inventory

While the likelihood and impact of various future conditionals were evaluated, the factor which was found to have the largest impact on the DPRK’s arsenal size was the operation of the ELWR. As the ELWR has shown signs of commissioning since October 2023, we have extended the model forward by ten years, taking into account two scenarios of its operation. Figures 7 and 8 show the forecasted impact of ELWR operation on the DPRK’s Pu and HEU stockpiles over the next decade.

How the North Koreans will use the ELWR is unknown. If it is operated similarly to a light water reactor, the fuel would be kept in the reactor for 2-3 years and produce plutonium containing isotopes that may require a more complicated weapons design. If the reactor is unloaded and refuelled more frequently, it could produce more plutonium and with an isotopic composition requiring less complexity in weapons design. If the reactor is stopped and refuelled more often, this would interrupt power generation, if that indeed is a purpose of operating the reactor.

We also do not have verified information on the design of the reactor, including its fuel type or thermal power rating. In absence of independently verified information we do have technical information provided by the North Koreans to a delegation that visited North Korea in 2010. The North Koreans may have provided misleading information, but in the absence of other information that could constrain reactor operations this is the starting point for understanding this reactor. VERTIC and the University of Uppsala partnered to model potential reactor operations which could then be used to build scenarios for DPRK fissile material production projected to 2034.

Two scenarios were developed: first that the reactor is operated as a pressurised water reactor (PWR), with three fuel loads in the ten-year period and secondly the reactor is optimised for plutonium isotopics suitable for less complex weapons and loaded and unloaded multiples times per year. A simplified model, not with use of Orion software, where enrichment is now allocated between enriching fuel for the ELWR or direct HEU production was developed.

Figure 7: Pu results from simplified forecasting model, 3.5% enrichment and 3 fuel loads over 10 years (PWR) and 1.5% enrichment with 4 fuel loads per year (WGPu optimisation).

Figure 8: HEU results from simplified forecasting model, 3.5% enrichment and 3 fuel loads over 10 years (PWR) and 1.5% enrichment with 4 fuel loads per year (WGPu optimisation).

Conclusions

By modelling the fuel cycle as a system we are able to see the impact of one area on another – for instance the impact of using centrifuges to enrich reactor fuel rather than being used to enrich HEU for weapons. Our scenarios suggest that the DPRK's current uranium supply is sufficient to supply a variety of enterprises without stifling the production of HEU. However, if the DPRK wanted to significantly expand its HEU enrichment capacity while fuelling new reactors, their current supply chain of uranium ore may not be sufficient.

The model also highlights that the DPRK’s plutonium production is currently not affected by the choices they make for producing HEU. While the 5 MWe reactor remains active as their main source of plutonium, steady plutonium production will continue. However, the 5 MWe reactor cannot operate consistently and safely forever, and replacing that source of plutonium may have implications for its capacity to produce HEU. We have not yet seen a strong link between the Pu and HEU pathways; all scenarios predict a similar range of plutonium. The reasons for this are twofold – first is that there is an unconstrained uranium supply, so after the first operational period we have assumed full reactor loads. The second is that even with scenarios such as using slightly enriched fuel for tritium breeding, plutonium production is approximately balanced by the expected decrease in neutron flux. However, this relationship between the Pu and HEU pathways may change with operation of the ELWR.

What are the chances that any of the scenarios presented actually captures what the DPRK has been doing? To cover this, we have looked at a wide range of options consistent with open sources. Perhaps we have been too conservative when filling in the gaps in our knowledge: we may have underestimated the uranium supply, the scope of the enrichment programme, or oversimplified the performance of the 5 MWe reactor. On the other hand, perhaps the DPRK’s uranium supply has been far more constrained - or their gas centrifuge programme encountered more significant difficulties - than we have modelled.

When extrapolating without supporting data, the possibilities become dizzying and the potential nuclear material stockpile varies considerably. Our approach has taken validated open sources, filled in the gaps with reasonable assumptions and attempted to match them to scenarios of fissile material production. We cannot know what the DPRK is actually doing without additional sources.

Finally, we should acknowledge the degree to which our results should be considered independent. As can be seen from our reactor and reprocessing operations plots, we have pulled in data from the IAEA, governments and other NGOs. We also studied estimates and assumptions in particular from the Center for International Security and Cooperation at Stanford University and the Institute for Science and International Security in Washington. In addition, we have drawn on analyses and supporting data from 38North, CSIS and others. In some cases, we departed from their assumptions and in other cases we found no reason to. Over time, as we learn more about what has happened in the DPRK we can test our assumptions and improve the estimates of this model.

Use the interactive map tool below to view the model’s estimates of fissile material accumulated at various stages in the DPRK’s fuel cycle from 1978-2023.

Credits

This work is part of an ongoing joint project between the Verification Research, Training and Information Centre (VERTIC) , the James Martin Center for Nonproliferation Studies (CNS) and the Royal United Services Institute (RUSI) , and is funded by Global Affairs Canada. We also thank the UK National Nuclear Laboratory (NNL) for use of Orion software.