Arms Control Wonk ArmsControlWonk


CNS made some beautiful 3D models of Iran’s IR-1, IR-2M, and IR-4 centrifuges for the NTI website. These weren’t easy to make. Fortunately, we have some smart, innovative, hardworking students. Bo Kim’s just finished her first year as a Nonproliferation and Terrorism Studies student at MIIS and is dividing her summer on campus between a CNS and a Cyber internship. Watch out future employers, she’s fluent in Korean too! Without further ado, Bo will explain how she took the measurements to make the IR-1.

Author: Bo Kim

Centrifuges, like the rest of us aspiring Instagram models, suffer from bad photo angles.

If you look closely, (or accurately measure as I will shortly) the ends of the centrifuges are smaller (i.e. further away) than the middle of the centrifuge, where the camera is positioned.  There are several great open source tools that correct pincushion or barrel distortion, such as Hugin or Gimp 2.8.  However, I found that Photoshop’s Perspective Warp function was more user friendly with a smaller margin of error.

Photoshop’s Perspective Warp Tool

1. Drop image into Photoshop and create a new layer via copy. (Precautionary measure.)  Layer > New > Layer Via Copy

2. Perspective warp the new layer. Edit > Perspective Warp.

A menu of options will appear in the navigation bar.  Click ‘Layout’ to see a box pop up inside the image. (shown in red below).

3. From here, move the four corner pins to match the face of the centrifuge.

You’ll already see here that the edges of the centrifuge are not perfectly linear. If you look closely, the holes of the bisected coils start out in line with the red perspective box, and moves off the line as the line progresses.

Then, hit ‘Warp,’ and the inner grid will disappear.

From here, click the ‘Vertical/Horizontal’ alignment button, and the centrifuge is now rectilinear.  You can see the warp on the side.

To reduce the margin of error, I photo warped both of the IR-1s in this image, plus as many IR-1s as I could find, and averaged the measurements.  After several nights of wanting to gouge out my eyes, I finally came up with numbers that added up.  Skepticism forced me to compare the measurements of the centrifuges with and without the perspective warp, and while no photo is ever going to be a 100% accurate, I’ve noticed that when building the 3D models, the measurements from the perspective warped photos seemed to be truer to the original images.

Web Plot Digitizer 

Now on to WebPlotDigitizer to measure the freshly warped centrifuges!  File >  Load  Image….

Choose appropriate file, then load.

For measuring images within photographs, I always go with Plot Type: Image.  Then ‘Align Axes.’

Then, at the top, ‘Analyze’ to ‘Measure Distances.’  From here, I hit ‘Add Point’ on the right every time I want to add a new measurement.

You’ll also see here on the right, that there is a zoom window.  You can use the ‘+’ or ‘-‘ buttons to zoom in or out.  The zoom window will move when you mouse over the image.  The intersecting lines in the zoom window helps pinpoint the starting measurement.

The first measurement I made was a horizontal line to define the floor.  The camera angle of this photo plus the color of the floor make it very difficult to define a flat surface from where we can start measuring.  Luckily, we can see in this photo the shadows of where two opposing ends of the centrifuge touch the floor.  I drew a line here from edge to edge to define my horizontal axis, as well as the hypotenuse of my base.  Simple math (including diagrams) coming up.

I then measured the height using Project Alpha’s calculation.

I measured from the top of the centrifuge to the defined floor.

Now for the simple math.

Project Alpha says our total height is       210 cm.

Webplotdigitizer gives a distance of        460.92 units.

So setting up my calculations:                     known = (unknown) (constant)

210 cm = (460.92) (C)

C = 0.456

This constant, C = 0.456 must be multiplied to all of the measurements in this image.  Like a numerical scale.

One last tiny bit of math.

The base distance is 77.72 units in Webplotdigitizer.  So multiply that with the constant and I know the length is 35.44 cm.

If you loved geometry (anybody?) then you’ll remember the hypotenuse of an isosceles triangle with sides x is √ 2 X.

So, that gives us:

√ 2 X = 35.44 cm

X = 25.06 cm

Seems about right.

Now for the rest of the measurements, I’ll just give you a quick snapshot of my final marked-up product:

Yes, I really did measure every nook and cranny on these centrifuges -for several images, of several centrifuges, several times.  I found it useful to do two separate images in Webplotdigitizer- one for horizontal measurements, and one for vertical.  If some of the measurements are ever hidden, or if you’d like to keep track of the data, you can hit ‘View Data.’


Once you have all your measurements, you can then move on to 3D modeling!

And if you start feeling nostalgia for the math, you can move on to next level centrifuging- counting the number of centrifuges in Natanz, geolocating the facilities, and then lovingly placing each centrifuge into a cascade.  50,000 centrifuges will guarantee you a hat tip from the Grand Guru herself, Melissa Hanham.  (Which will likely be immediately followed by requests to recolor, rearrange, recount, remake, and rerecord everything.)[<--MH Edit: By Grand Czar Jeffrey Lewis]

Happy centrifuging.



I just got back from London, where Nick Gillard and I tried to stump the very smart participants of RUSI‘s UK PONI conference with a pub quiz. Despite a few technical glitches and a few fire alarms, it came down to the sudden death tie-breaker question. Many thanks to Andrea Berger and the whole RUSI team for indulging our madness with iPads and pub food for the teams!

See if you can do it too! Remember they only had wifi enabled iPads, and these questions were projected on a screen. The winning team enjoyed a 50£ tab at a local pub thanks to Nick! I can only give you ACW fame… Enjoy!

Question 1:

Answer this question in 2 minutes or less!

Question 2:

Answer this question in 5 minutes or less.

Question 3:

Answer this question in 5 minutes or less. (We had to clarify that offshore meant water not air)

Question 4:

Answer this question in 2 minutes or less!


Question 5:

See if you can answer this in 5 minutes or less.


Tie-breaker Question:

First correct answer won the game, see if you can do it in less than 5 minutes.



Sadly, there is no textbook for imagery analysis. I would love to contribute to one someday, but until that becomes an option, here’s the March 1996 Unclassified Photo Interpretation Student Handbook. I picked it up from from a rather unusual meeting on a DC trip and several students lovingly/begrudgingly scanned it for our benefit. It’s an excellent resource for identifying everything from transportation to military installations. You’re welcome!

Searching for silos? 

Sizing up subs? 


Seeking sulfur? 

Still not enough Photo Interpretation? Then check out the 1944 vintage stored at the University of Nebraska. It’ll help you verify vessels!



A lot of people are asking how to take measurements and make 3D models from 2D images. If conditions are juuusst right, SketchUp’s Match Photo technique is the way to go. Unfortunately, in our world, conditions are almost never just right, unless you happen to be able to take the photo yourself. Here’s an alternative method using Web Plot Digitizer, which is still free and can scale for as much (or as little) information as you have (keeping in mind a greater margin of error).

A few days ago the DPRK’s state news agency, KCNA released the first images of North Korea’s new submarine launched ballistic missile (SLBM). These photos we’re great, but with just sky and water, there wasn’t much context about the size of the missile.

That is until one of our eagle-eyed research assistants, Dave Schmerler, spotted this:

BOOM! and, we were off running (well, typing furiously on a JFK-> SFO flight). Surprise! That carefully crafted message of Kim Jong Un hanging on his yacht smoking a cigarette and watching a missile launch from a submarine? Not so much. Turns out they start off launching from an underwater platform just like the rest of us.

So, to the measurements:

1) Geolocate the ship! 

Dave already took care for that for us. It’s in Sinpo! (40.026008°, 128.166174°) Using the measurement tool in Google Earth, we can see that from stern to cabin, the ship is about ~10.03 meters.


2) Load the screen shot from YouTube into Web Plot Digitizer 

Load in the very best screen shot you can muster from the YouTube video which shows both the missile and the ship. Next, choose the “Map With Scale Bar” option, and hit “Align Axes.” So… we don’t actually have a map with a scale bar, but we are going to simulate one. And, because the entire ship isn’t in the image, we are going to use the distance from the stern to the cabin as our scale.

Zoom in as much as is comfortable, and use the window in the top right to place your points. You can adjust them as needed. I measured along the top of the water to keep things level. When ready, click “Complete!” and enter the units. I measured 10.03 meters on Google Earth, so that’s what I put in.

3) Measure! 

Under the Analyze tab you will find the “Measure Distances” option. Select it and make sure “Add a Pair (A)” is highlighted. Again, zoom in as much as is comfortable, and use the window in the top right to place your points. If you want to start over you can “Delete a Pair” or “Clear All.”

I did this several times, and got slightly different measurements, but they were all around 9 meters long with 1.5 diameter. There is a kind of faint halo that forms around the missile, and I tended to discount it in the measurements. I’d be interested to see what measurements others get, so that maybe we could average them. This rough estimate makes it seem like they are working from an R-27.

Using a higher resolution image of the missile (which, unfortunately, doesn’t have the ship for scale) I got a ratio of  90.64 to 14.95 pixels.


Hey! Dave even made a great 3D model for NTI! Check it out:

4) The Fine Print

Like it says in the title, these are just back of the envelope calculations. Please don’t set your ballistic missile defense to it ;)

First and foremost, these aren’t very high res images. And, taking a screenshot of a YouTube video of a still image, is kind of scraping the bottom of the barrel. A pixel here and a pixel there on a low resolution image can add up. On top of that, we don’t know the distance or heading of either the ship or the missile relative to the camera or each other, which causes a greater margin of error.

Another fantastic research assistant, Bo Kim, will write up a more detailed account of how to use this process to measure Iranian centrifuges taking into account their angle relative to the camera. STAY TUNED!


One of the easiest and most useful methods for an open source analyst is to extract metadata from imagery.

Metadata is data that is often included with an image, such as the time it was taken, the type of camera that was used, and yes, if you are lucky — GPS coordinates. This data is useful for photographers, those who like to stalk cats, and people like us: geolocators and myth-busters.

There are many different types of metatadata, some generated by cameras or phones, and others generated by editing software like Photoshop. Extracting metatadata can not only give you clues about when and where the image was taken, but also about whether it was altered. Last, if you are trying to generate a 3D model from a 2D image, it can give you some clues about the distance the object is from the camera.

Last week I posted an image to, challenging fellow wonks to geolocate the image. There were some grapevines and California poppies in the picture, but if you hadn’t been there before, it would have taken some time to find the location. That’s why your first step should always be to check for metadata.

Photoshop is a good tool for extracting metadata, but if you are looking for free options, there are plenty. Try copying the above image’s URL and pasting it into one of the below:

Jeffrey’s Exif Viewer (No relation, I presume)

This site has some nice extras, though they are not always accurate. It will try to give you an approximate address, a color histogram, and a circle of confusion if possible. You can see that the geolocation is close, but not quite accurate. Also, the viewer shows the photographer facing the wrong way.


This site is cool, because in addition to the regular metadata it will give you an Error Level Analysis (ELA). ELA lets you see when the compression rate of the image changes. So if something is “airbrushed” or pasted, stretched or cropped,  you can see a difference.

No obvious alterations here, but you can see a faint grid pattern (cross your eyes like you are looking at a stereogram) showing that the jpeg was likely resaved.

Why worry if an images is altered? Because: Russia, North Korea, Iran, Burma

Fortunately for us, nothing has replaced the wonk yet! As helpful as the metadata from the geo quiz photo was, it wasn’t 100% accurate. It will put you within a few meters of where the photographer was standing, but it’s still best to go to Google Earth to fine tune the coordinates and the direction of the camera.

Ah yes, now imagine a glass of chardonnay in your hand!


Back by popular demand: a geo quiz! This is a teaser for a how-to I will write next week.

Q1: Where is the photographer standing?

Q2: What is the address of the nearest building shown in the picture?


Put your answers in the comments below, and NO PEEKING!


Chemical Weapons Dumped at Sea Google Map

Many people ask where they can get satellite imagery. Working at a nonprofit, my preferences tend towards the free, but there are some great resources on the cheap too. Why check more than one map? Because: Volkel.

Here’s a list of my favorites:

Google Earth

It’s free. Even Pro is free. It has a time slider, allowing you to look at change over time easily. You can also overlay images (or maps!). There are a ton of KMZs available for download. Some of these may be curated by experts, others are wikis, you can even make your own! Last, it’s 3D. You can import your own 3D models, calculate elevation changes, even build a missile flyout!

Google Maps has some of its own advantages — it’s easier to embed being the one. You still need to use Google Earth to see the date of the image. You can also set alerts for imagery updates, and there’s even 3D data!

Bing Maps

Bing is getting better — for example, there’s road data for South Korea now (yikes, that took a long time). The greatest advantage over the maps below is this little tool, which tells you the date of the imagery.


HERE (formerly Nokia Maps) did a massive upgrade to its imagery last year. It’s still primarily best for the US and and Europe, but you can make some great finds.


Try Yandex and Baidu. Many draw from the same DigitalGlobe catalog, but may be from a different date. If you are interested in South Korea, try Daum and Naver, but they won’t help you much north of the DMZ.


This isn’t the high resolution imagery you associate with the above, but depending on your needs, it might be better. The USGS’ Earth Explorer tool lets you search an enormous database of free or low cost imagery, including declassified images and Landsat’s thermal infrared bands.

Corona Atlas of Arkansas

I actually just learned about this tool to search Corona imagery thanks to @arawnsley. I will be using it in the future!

Digital Globe Foundation

Still don’t have what you want? If you are affiliated with a university, you can apply for an image grant.


The sky is falling!

The United States is about to knock over the first domino and start the nuclear proliferation chain reaction in the Middle East. With the region at war, the argument goes, the American attempt to negotiate with Iran over its previous nuclear weapons work (and current nuclear infrastructure) is painfully naïve, and portends a future of nuclear-armed states. This new nuclear future, we are led to believe, will begin in Saudi Arabia; be followed by a Turkish nuclear weapons program; include a Hashemite bomb in Jordan; and end with Cairo dusting off those Nasser era plans for nuclear weapons (Hey, with Sisi going full Nasser on us, perhaps he may approach China about purchasing a weapon.)

One problem with this argument: It is at odds with all that we know about nonproliferation decision-making and, at least in the case of Turkey, Jordan, and Egypt, is near impossible, given their choices about nuclear financing. The region is a mess. A total mess. But proliferation is – thank god – not something that should dominate the debate. In fact it misses the point entirely. Folks concerned with nuclear issues in the Middle East should certainly keep an eye on proliferation concerns, but should focus on a more serious issue: nuclear safety.

From the little we know about Saudi Arabia’s current nuclear plans, the Kingdom has pursued a rather benign approach to nuclear energy. After threatening to “go nuclear” for years, the Saudis chose to sign a MOU with South Korea’s KEPCO. Just next door, KEPCO is putting the finishing touches on the first of four reactors at the Barakh power plant. KEPCO has a great reputation for finishing projects on time, but is not really known to be the Korean peninsula’s equivalent of the AQ Khan network. Does South Korea have a pristine nonproliferation record? No. Has it exported enrichment or reprocessing technology to a customer? No. Will Saudi proliferate using a KEPCO APR1400. Hell no.

There is an elephant in the room. Pakistan, the argument goes, has reserved some warheads for Saudi Arabia’s defense. The Kingdom subsidized the Pakistani nuclear weapons program and therefore could try to collect on its investment. There is simply no way to account for this, other than look to history for some examples. The aforementioned Nasser asked a few countries for nukes during the 1960s. They said no. The Saudi case may indeed be different, owing to the Kingdom’s alleged investment in the program. My colleague and friend Philipp Bleek tackled this issue a few months ago, arguing:

History suggests that while some states have trumpeted their potential desire for nuclear weapons—think Germany in the early years of the Cold War, or Japan more recently—they tend not to be those that later went on to actually acquire them. And for good reason: calling attention to proliferation intentions is counterproductive if one is intent on actually proliferating. Instead, states tend to draw attention to their potential proliferation in the service of another goal: rallying others to address the security concerns that are motivating potential proliferation, and especially securing protection from powerful allies. At least one state, though, trumpeted its potential proliferation while actually pursuing nuclear weapons, and the case is an instructive one for analyzing Riyadh’s recent nuclear saber rattling.

Again, the Kingdom chose KEPCO.

Turning to Turkey, Egypt, and Jordan – or as I have dubbed them, the Rosatom three. These three countries don’t have the cash to pay for a nuclear power plant. Or, if they do, they don’t want to pay ~$20 billion up front for their 70 year investment. To pay for their reactor, all three rely on a financing scheme known as Build, Operate, Own (BOO). Turgut Ozal came up with this idea in 1983 to help Turkey develop, after the country switched from import industrial substitution to export oriented capitalism. The BOO model was meant to attract private investment to what Ozal dubbed Turkey’s most trust-worthy sector: state-owned energy utilities.

This model is the reason why Turkey failed to procure a reactor between 1983 and 2010; more specifically, Ankara failed to give the vendor a treasury guarantee. It still doesn’t, arguing that the power purchasing arrangement with the electrical utility is good enough. Enter Rosatom. The Russian state-owned nuclear firm is not a private entity and receives ample funding from the Kremlin. Rosatom’s reactors are one of Russia’s few high-technology exports and the industry helps to employ Russia’s legions of Cold War era – and current – nuclear experts. The Kremlin, in turn, has also sought to use the company, as a tool of foreign policy, in much the same way Gazprom has become an arm of the Russian MFA.

For these reasons, Rosatom has embraced the BOO model – despite ample evidence that the financing for such a project makes little sense for the nuclear vendor. Under the terms of the BOO model, the vendor agrees to pay for the cost of construction and operate the plant in perpetuity, in exchange for the host-country to purchase a fixed amount of electricity at a set rate. For Turkey, that rate is 12.35 US cents per kilowatt-hour for 70% of the power produced at reactors one and two, and 30% of the power at the same price from reactors three and four.

After an agreed upon time – usually 15 years – the vendor is expected to have recouped its investment. In turn, Rosatom will then collect a percentage of the profit the Turkish state will make from the continued sale of Russian nuclear energy for the lifetime of the nuclear plant (70 years). This financing model puts pressure on Rosatom to finish the project on time, or otherwise risk extending the time in which it will recoup its initial outlay of expense – keeping in mind that the Kremlin is footing the bill for a $20 billion dollar investment. The Turkish reactor is now 4 years delayed.

Jordan and Egypt have announced that they too will pursue the BOO financing model. The good news: This model makes it all but impossible for the Rosatom three to proliferate. The reactor – a VVER-1200, or VVER-1000 – will be Russian owned and operated. The spent fuel pond will be Russian owned and operated. The fuel, Russian supplied. You get the point. So, if the Rosatom three wanted to proliferate, they would have to steal spent VVER LWR fuel from a Russian owned spent fuel pond and then reprocess it in facilities that don’t exist. They would have to do this without getting caught. Good luck.

The bad news: The BOO model is a regulatory nightmare. To ensure that Rostaom does not cut corners to hasten construction, the Rosatom three need strong, independent nuclear regulators. The regulator must also be empowered to influence the behavior of the operator. How will this work with a foreign owned and operated plant? Is Rosatom beholden to Turkey, or Moscow. At a bare minimum, you need a strong regulator and clear guidelines. The evidence is not encouraging: Turkey’s regulator is weak and beholden to the Prime Ministry for all its funds. Egypt’s is geriatric. And Jordan has no nuclear history. Not good.

If you add Saudi Arabia and Iran’s Bushehr, you can start to see a “nuclearized” region with little nuclear expertise – and nuclear regulations that are very much in their infancy. Is this the end of the world? No. There is still time to work through these upcoming challenges. However, the regulatory issues are real and the international community has an incentive in ensuring that the region “gets this right.” Focusing on the long disproven idea of nuclear proliferation chains is not helpful and actually distracts from the issue we all should be focusing on: nuclear safety in a region (mostly) using creative financing techniques to “go nuclear.”


I’m deviating from my original plan to blog primarily on open source resources to talk about an issue I usually shy away from. It’s easy to be pigeonholed as the WOMAN rather than the expert. I also like to compartmentalize my work and home life. In a venue where I’m teaching people to use open source analysis, I don’t want to invite attention to my family. Too many women get their lives ruined on the Internet these days.

Last, I hope nobody thinks these comments are meant to undermine friends and colleagues. There’s just a hot pink elephant in the room. Let’s talk about it.

5:30am: wake up and luxuriate in the fact that I have slept in. (Jeffrey knows what I am talking about). I’m a new mom. This quickly turns to guilt because my new twin boys are at home with my husband and I am not. I’m in DC on my first big trip away from them. I haven’t traveled until now because I am a clingy first time mom, and more pragmatically because I was breastfeeding. There’s no place to plug a cellphone in at a big conference, let alone a breast pump. That’s not to say I stopped breastfeeding to come to this conference, just that I couldn’t have attended until now when I naturally had to wean off.

6:45am: arrive at conference venue to start up computers on our table display. We’re promoting the 3D modeling work that I am a big part of. It’s great to show off the work, but I am disappointed that almost no other staff members have signed up to take a turn at the table during the breaks. I complained to a colleague a few days ago that I thought she would have at least signed up for a turn. We are close friends! She delicately explained that she didn’t want to sign up until at least one of the men attending the conference signed up.

7:30am: attend the Women of Mass Destruction session. Two years ago, I sat next to Cheryl Rofer and struck up what I hope will be a life long friendship. She was a chemist at Los Alamos, and in her “retirement” runs the  Nuclear Diner blog. We’re facebook friends now. Sometimes she gives me technical analysis and sometimes she mails hand knit caps for the babies.

Cheryl is now my ally, and I can talk to her about stuff that still can not be said out loud at the W(oman)MD session. Even though it’s a safe space, in the age of Twitter, it’s still on the record. Nobody, particularly the younger crowd, wants to be perceived as a “problem.” Problems are not very employable.

So predominantly, the questions surround non specifics. How do I respond when men interrupt me? How do I find a job? How do I get access? How do I get taken seriously?

11:00am: get mistaken for an intern while standing behind the display table.

3:00pm: try standing in front of the display table with more success.

4:30pm: engage in a Twitter discussion about the gender of the speakers and moderators on the conference’s panels. #manels is a hot hashtag these days. It’s an easy metric to throw around and a hard one to do something about.

The conference has had a great deal more women on the stage this year, but as a percentage it’s still low. One colleague pointed out that it’s the highest its been at 30%, however others argue the number is really lower due to the fact that many of the female speakers are in the optional side meetings and some speak on more than one panel. There could be a lot of reasons for low participation. Perhaps women are still rising through the ranks, they don’t feel confident in their expertise, or they have a tough time leaving their second job as caretaker behind.

One of my colleagues thought the organizers should do more. She argued that even 30% representation was not enough for congratulations. The next day, I was approached in person and chastised by someone who thought her comments were offensive and inappropriate for Twitter. Also, he told me, three female speakers cancelled last week. I wish I had thought to tell him that there would probably be less tweeting if there was a better way to participate in the conference, but I didn’t. I passed on the message.

10:00pm: sing Baby Beluga into a cellphone.

Women are different — from each other and from men. This is a tough conversation to have, but without women’s opinions we are losing half the stakeholders, and half the potential solutions.


Catherine here.  For the past year, Jeffrey and I have been looking at China’s Korla Missile Test Complex in Xinjiang, which is where we believe China conducts test launches of its hit-to-kill interceptor.  Jeffrey put up a brief post about part of the site in August, after China conducted a missile defense test this summer on July 24, 2014. The US State Department characterized the event as an anti-satellite test, but Jeffrey likes to point out that it’s better to call it a hit-to-kill test.  What it kills isn’t so important.

In this test, as well as tests on January 11, 2010 and January 27, 2013, China has reportedly launched the HTK interceptor, usually called the SC-19 in the US press, from a site near Korla (库尔勒市), in Xinjiang province.  A possible test occurred on September 25, 2010, but was not officially acknowledged. In that past, China used a CSS-11 missile launched from the Jiuquan Satellite Launch Center as a target. (Side note, the US intelligence community calls Jiuquan “Shuangchengzi”, which is where the SC in in SC-19 comes from.)

We can now say with high confidence, based on some open-source research, many things about the Korla MTC including the location of many of its assets, that it is subordinate to the General Armaments Department (GAD) and the location of a number of previous missile defense tests.

The main base is located at 41°42’37.76″N 86°11’22.69″E. The hit-to-kill launch site is located at 41°32’14.18″N 86°21’11.82″E.


General Armaments Department in Korla

We were able to identify these bases—and link them to the GAD—with a little help from our colleague Iain Johnston, who sent us an email after the July post:

I enjoyed your Arms Control Work piece on tracking down the Korla test site. You mentioned in the post that you didn’t know if it was a GAD site. Perhaps you’ve already tracked this information down, but FWIW I found a website that lists GAD bases. There is one at Korla and it is identified as the 63618 unit. I also found a reference to this unit in a local government webpage for the Bayinguoleng Mongolian Autonomous Prefecture in Xinjiang, within which Korla is situated.  The unit is identified as being situated in the outskirts of Korla city. Here’s a video from the unit made in 2010 to celebrate soldiers leaving the force. The credits at the end note that the movie was made by unit 63618 in Korla. The webpage of the Korla city government also refers to this unit. Judging from the topics of articles on CKNI that are co-authored by scholars affiliated with unit 63618, its work appears to be related to satellites and missiles (e.g. “Location Model and Error Analysis of Space Target Location by Early Warning Satellite” in Aerospace Electronic Warfare; “Detecting Probability Calculation on Moving Space Target of Space Detector” in Electronic Information Warfare Technology; and “Research on the effectiveness of early warning satellites to detect early warning indicator system” in Aerodynamic Missile Journal.

Unit 63618 appears to be subordinate to the Jiuquan Satellite Launch Center from the list of GAD bases Iain provided. Even without the GAD listing in Chinese, one can infer that Korla falls under Jiuquan. The ever-useful Directory of PRC Military Personalities lists the Jiuquan SLC as Base 20 (63600). The next base is Base 21 (63650). The Korla unit—numbered 63618—falls logically under Jiuquan.

The available information confirms that Korla is subordinate to the General Armaments Department, which is responsible for developing things like hit-to-kill interceptors, and that personnel at Korla work on targeting things — missiles and satellites — in space.

The Unit 63618 alumni video, though, was the real goldmine and crucial in geo-locating the unit. Videos bidding farewell to rotating personnel or remembering the anniversary of a particular unit often appear in internet searches. This type of media can provide helpful context of the environment in which a military unit spends its time and the necessary daily tasks of life it undertakes.

Screenshot of the Unit 63618 alumni video title.

The Unit 63618 video contains ground-truth images of both the main base and the hit-to-kill launch site. The most prominent features of the main base to match to the available satellite imagery include the main gate, the back of the building, and the base’s basketball court. These images provide visual confirmation of the location of the KMTC.

The main gate of the KMTC.

Basketball courts at the KMTC.

Administrative building at the hit-to-kill launch site.

Construction at the hit-to-kill launch site.

The alumni video, made in 2010, clearly shows the hit-to-kill launch site under construction. This timeline also fits with the satellite images of the site, as well as the general development of the SC-19 interceptor program that began at the KMTC with the 2010 missile defense test.



One thing we wanted to check was whether an interceptor launched from the KMTC could hit a test missile launched from the Jiuquan SLC, which is how the press reported a previous intercept test. David Wright from the Union of Concerned Scientists kindly modeled the test for us and analyzed the trajectories of the target and the interceptor vehicles (You can read a more thorough analysis on the test from David here).

Two trajectories for both the interceptor (blue) and target (red) missiles. The dots show 30-second intervals, with the target missile launched at t = 0.

Based on known and assumed parameters of the CSS-X-11 and SC-19 vehicles, David is reasonably confident that the intercept is possible and matches known timelines of the test.

At this point, we’re probably polishing the cannonball, but we acquired satellite imagery of the KMTC on the day of a known missile defense test.  That image also confirms the base’s involvement in missile defense testing.

Through a generous imagery grant from the DigitalGlobe Foundation, we have a satellite image of the test facility on January 27, 2013. We weren’t lucky enough to capture the TEL on the launch pad, but there are many visible features that show preparations for an interceptor test.

The hit-to-kill launch site on January 27, 2013. Image © DigitalGlobe.

The launch pad has been cleared and there is significant vehicle activity at the facility, which is normally unused except during periods related to testing.  Most important, the instrumentation sites have trailers in them, something that only occurs for tests.

The administrative facilities and launchpad on the day of the January 27, 2013 test. Image © DigitalGlobe.

We checked with a few missile defense experts. Here is a summary of the signatures they found interesting:

  • Near launch-photo documentation
  • Short range radar tracking
  • Telemetry reception for pre-launch and early flight testing
  • Tracks caused by moving non-permanent generators and other instruments in support of data recording
  • Data reception and communications equipment
  • Cables for instrumentation and remote recording
  • Dishes 3-6m in diameter for telemetry
  • Radio antennas and radar
  • Mini-dines for sheltering optical equipment
  • Cameras

The available satellite imagery strongly suggests the facility is active in HTK vehicle testing activities.


LPAR near Korla

Last thing! There is also a Large Phased Array Radar (LPAR) facility between the main base and the test base. Sean O’Connor identified the LPAR in 2009, located at 41°38’30.46″N 86°14’15.27″E.  Obviously, an LPAR would be useful for looking at missile intercept tests.  As you can see from the image, it turns.  We weren’t able to get a picture of the LPAR on the day of the test — the picture cuts off — but that’s a signature for future use.

In the left image, taken shortly before the January 2013 test, the LPAR is angled differently than in the right image, taken in October 2013.



The satellite imagery, alumni videos, and listings of GAD units strongly indicate that the Korla Missile Test Complex is the test facility for HTK technology. The question of whether the tests at the KMTC are for missile defense purposes or ASAT development is far trickier to answer, as the HTK vehicle can be utilized for either purpose.

This geo-location exercise raises a larger issue to which to return in future posts—although China is not very transparent in releasing official information on military programs, this is not the case when it comes to the ubiquitous information available on Chinese social media information platforms. Much incidental information, most of it mundane and about ordinary life, exists and helps to inform a broader picture of Chinese military capabilities. Of course, analysis of Chinese social media information requires diligence and caution, and review of official releases requires considering whether information was released as a part of “selective transparency.” But at the very least, policy experts should consider helping government-to-government exchanges to catch up to the transparency present in social media. In the case of missile defense testing, a reasonable step might be for China and the United States to exchange notifications of upcoming tests.