CISAC Team Releases Satellite Imaging-Based Study of North Korean Uranium Mines

satellite

North Korea currently has only one publicly known uranium mine—the Pyongsăn uranium mining and milling complex—that serves as a first step in the country’s pathway towards nuclear weapons.

Using a combination of multispectral imagery sourced from the European Space Agency’s Copernicus Sentinel-2 satellite and a review of geological analyses dating back to 1955, a new study from Stanford’s Center for International Security and Cooperation (CISAC) in Jane’s Intelligence Review by geological sciences postdoctoral fellow Sulgiye Park (PhD ’17) and CISAC honors student Federico Derby (BS ’19) looks for evidence of uranium mining in North Korea, going beyond what is currently available in open sources in order to estimate the uranium resources and their locations in North Korea.

The peer-reviewed CISAC study has identified around 18 additional sites in North Korea where the hyperspectral signatures and geological profile combine to suggest the possibility of uranium mining. Nevertheless, CISAC and Jane’s stress that the presence of these ‘hotspots’ does not imply the presence of an active uranium mine or related facility, but rather a site that warrants further analysis.

In this Q&A with Katy Gabel Chui, researchers Sulgiye Park and Federico Derby discuss their work on the project:

How did you land on this project? What made you think to look for more mining sites?

Sulgiye Park (SP) and Federico Derby (FD): Very little is known about the front-end of North Korea’s nuclear fuel cycle, particularly when it comes to the mining and milling processes of uranium production pathway. To date, assessments of this portion of North Korea’s nuclear fuel cycle have been mostly conducted through traditional (electro-optical) satellite imagery observations---the type of imagery that you can access through Google Earth, for instance.

We wanted to get a more complete grasp of North Korea's uranium mining and processing capacity by conducting a multi-disciplinary approach that combines both the visible signatures from multi-spectral satellite imagery and a geological dataset that contains information such as mineralogy and geochemistry. The two individual methods come together at the end to provide information that encapsulates the potential regions likely to host uranium deposits and mines.

What is multispectral imaging? How would it ordinarily be used, and how did you use it for this project?

SP and FD: Traditional electro-optical satellite imagery exploits only three portions of the electromagnetic spectrum; namely, the blue, green and red bands. In general, when using the term “multispectral” within the satellite imagery community, we are usually referring to a satellite system that covers a few to tens of different bands in the electromagnetic spectrum.

Multispectral imagery is used in a wide variety of industries, to measure things like water turbidity, crop healthiness, vegetation quality, etc. For this project, we focused on using spectral fingerprints. Basically, every object – whether it be a mineral, a living thing, water, etc. – has a(n in theory unique) spectral fingerprint. Spectral fingerprints are measured as the intensity of the object’s reflectance of light at a specific wavelength. Varying across wavelengths – hence the importance of having a multispectral system that can give you access to different ranges of the electromagnetic spectrum – you ultimately get a spectral curve that is unique to the item you are studying.

The spectral fingerprints you collect on a specific image can be compared to previously collected fingerprints stored in what is usually termed a spectral library, for classification purposes. Basically, if my spectral curve of a given pixel (or set of pixels) looks super similar to that of gold (for which I obtained a reference spectral curve from a spectral library), then it is probably gold. Obviously, this matching is performed in a more rigorous manner, but you get the idea of how the process works.

In this project, we used the Pyongsan uranium mine in North Korea (arguably the only well-identified uranium mine in the country) as my reference spectral curve. Essentially, using various imaging techniques, we traversed North Korea looking for pixels whose spectral curves are similar to that of the Pyongsan uranium mine. Those are the ‘hotspots’ we identified.

What most surprised you in both your work and your findings?

SP and FD: The fascinating match between the 'hotspots' identified through satellite imagery analysis and the geologic information available in maps and reports. The majority of the 'hotspots' appeared adjacent to the limestone formation from the Ordovician period (circa 445-485 Ma) that are in contact with a specific sedimentary rocks of upper Proterozoic group. Part of the geologic characteristics of the 'hotspots' regions were similar to what had been observed in the Pyongsan (the most well-known) uranium mine of North Korea.

What was most surprising in the work itself? What was difficult in doing the work?

SP and FD: It was surprising to see how much we still don't know about North Korea despite the amount of effort that had been invested. There is no consensus reached regarding the location and the total number of uranium mines in North Korea.

One of the bigger difficulties we had was finding credible geological data and information.

What is the one thing you think someone should take away from your study?

SP and FD: That there are still many unknowns. While our study identified multiple regions with spectral signatures similar to the uranium tailing piles at Pyongsan, verification of uranium presence is still needed.

What are you working on next?

SP: I am still working on using a geologic approach to glean information on the uranium mines of North Korea. The further evaluation aims to identify a qualitative upper limit of uranium ore grade (quality) and quantity pertaining to all the suspected uranium mines in North Korea.

FD: I co-founded a startup focused on developing deep learning models for credit risk analytics (in Latin America). However, I will still keep in touch with my CISAC peers!