No, China Isn’t Punishing North Korea

A few weeks ago, some colleagues new to satellite imagery analysis claimed that China had drastically reduced trade with North Korea to punish Pyongyang for its most recent nuclear test. They made this judgment on the basis of a small number of satellite images of the customs area in Dandong, China and one image of the customs area across the river in Sinuiju, North Korea.

Small samples can be problematic. All sorts of things can go wrong — this is one crossing point, on one day, showing one mode of transport.  Some days are busier than others. What if the picture is from a weekend?  Or a holiday?  In this case, it turned out to be both — the “after” pictures were from February 13-14, the Saturday and Sunday at the end of the Chinese New Year.

Measuring something like trade by monitoring traffic flows is tricky.  It requires lots of images.  One has to consider the modes of transport for various goods, the entry points, and seasonal variations in trade. And weekends and holidays.

In other words, the  correct tool for the job is not a small number of high-resolution images, but a larger number of moderate resolution images from a company like Planet Labs.

As a researcher, I have access to Planet imagery for the purpose of providing expert analysis. I did a review of the border imagery and shared it with Planet. Paul Boutin then turned into a story on their Medium channel, replete with commentary from Steph Haggard and Andrea Berger. The picture at the top of the post is a screen shot of all the images stacked on top of one another.

Although these 2-4 meter images aren’t quite good enough for vehicle counts, you can definitely see whether the customs areas are busy or not. I am agnostic about whether trade is up or down, but the satellite images are not consistent with the idea that China is inflicting much pain on North Korea.

I don’t have much more to add.  Paul did a great job writing it up.  The one thing I do want to share is a note on method.  Paul mentions that I looked at ten good satellite images from February through July. Here is my list of the images and my assessment of whether the lots in Dandong and Sinuiju looked busy or slow.  Busy means they looked more like the Digital Globe images from January 15 and March 13, 2015; slow means they looked more like the Digital Globe images from February 13-14, 2016.  

There were one or two judgment calls, but with the Scene ID and a Planet account, you can replicate my analysis. There are so many pictures, though, that I think the result is pretty robust — which is the value of having lots of good pictures instead of one great one.