Open-source satellite data to investigate Xinjiang concentration camps

The second part of this series discusses techniques on how to analyse a dire human rights situation in and around Xinjiang’s re-education and detention facilities.

A statistical model was used to count the area of factory buildings (green overlay) inside a 1-mile radius around two of Xinjiang detention centres based in Onsu, Aksu

A pressing need to investigate characteristics of Xinjiang’s detention camps

Mounting number of facilities

List of ‘expanded camps’ extended

Screenshot of the expanded data list for expanded facilities (61 sites, in pink, some added features)

We will go through various ways to characterise/investigate facilities and their surroundings

Comparing 176 facilities for which we have the distance to residential buildings and changes in security features and specifically in whether towers got added to the site

Blue-roofed factory buildings

Images: The Bugur camp, at 41.808855284.3005783 (where the little white dot mark the camp in the upper part of the image), is a re-education facility in Bugur with ‘with considerable room for expansion’. It’s fenced inside and has a large external wall. ASPI expects “considerable room for expansion”. It makes sense there are three large quadratic areas in the northern part that are unoccupied. With a total size of 0.35m2, the camps itself is fairly large. ASPI counts 8 factory buildings within the compound. It’s unclear whether that’s including the one in the middle.
Most of the last available satellite images only cover the start of the pandemic (March, 2020)
Showing a grouping of the final dates (grouped by year) for open-source satellite images on Xinjiang camps
Left: the site when it was up and running, in 2019. Right: Blue factory building existed first in early 2017
Google Earth (left early 2018, right: July 2020, overexposed to see the blue buildings): With the Karakax list and details for the respective town, a new big factory complex was built (the pin in the south, 37°13’42.55″N 79°44’14.44″E). Blue-roofed factory buildings were erected next to the southern internment camp near the industrial park. Often the blue roofs are placed inside the camps’ compounds. Others are just very close by, e.g. across a road or within the same industrial site. Zenz writes that “just north of this compound are large new residential areas that appear to be part of a 380 million RMB poverty-alleviation employment and resettlement plan”
Higher security re-education camps and detention camps tend to be in closer proximity to industrial sites. It’s concerning in the light that many Uighurs are estimated to be subjected to forced labour practices
Revealing what’s inside of the blue-roofed building
Various views of the Dabancheng prison camp facility

Among the around 60 camps that have expanded recently, half of it are tier 3 or tier 4 facilities —detention centers and prisons with high security features.

Development of a Tier 3 prison camps Sanji in the north of the province
From barren land to a fully functioning prison facility in less than six months (coordinates here)
High-security facilities that expanded/changed with distinct characteristics that emerge elsewhere

What parking lots tell us about the camps during Covid-19

Before the height of the pandemic and during
Results 10/24/2019
Results for 3/9/2020

Calculating rooms and capacity

Different views of a camp’s location allows estimating the number of rooms and how many people it can host
Vanishing cars in the camp in Cherchen County ( 38° 6'16.85"N, 85°34'32.25"E). Whether all the inmates were moved is certainly tricky to answer. The images show little evidence that all the operation stopped. There are still cars within the complex, just not on the car parking lot.
Cars at the tier 2 facility near Payzawat (39.538372, 76.713606), decreased as the pandemic hit (timeline from left to right). It’s a Tier 2 re-education facility in Peyziwat, Kashgar with an external wall

Comparing camp sizes

There are clearly more facilities that have a few buildings, but some are considerable larger (sample of 231 facilities). Tier 4 high-security prisons tend to have fewer buildings.
10 of the largest camps — with Dabancheng listed twice as it got expanded massively. In 2018, the number of building counted was with around 40 buildings on site comparable to the facility in Onsu (#4). In 2019, construction efforts added nearly 30 new buildings. In 2020 satellite images show around 92 buildings, with construction still continuing. The Dabancheng facility (#1) also houses 16 factory buildings. With what researchers learned about the camp’s organisation, the odds are considerable that the factories maintain forced labour. The camp itself, so far, was not visited by international delegations/press/human rights investigators (acc. to ASPI). If China opens up and agrees to visits and inspections from human right delegation, this is the place to start.

Staking out camp size:

Using the ruler tool to stake out the camps

What can walls and towers tell us?

Left: Image of the entrance gate of the Dabancheng facility in 2018 (4 September 2018, by REUTERS); Right: walls and towers from outside, showing the perimeter fence of the facility (by Thomas Peter/Reuters — also 2018 as the high towers were removed after)
Finding the height of the tower structures in March of 2019.
Height of original towers where 14.5m hight.
Checking the height of camps’ walls and validating it via images taken on the ground
We run the findings from the Hotan camp against this one in Kashgar and see that the Hotan walls and towers appear to be higher. The buildings in the background are also distinct

Buildings shapes/outlines and location of camps

Detention facilities tend to be kept further away from residential buildings, but the results are inconclusive for high-security Tier 4 prison facilities

Agriculture/fields around the camps — investigating forced labour by detainees

Left: in 2018, right, 2020
Sentinel 2 images: false colour image from 09/21/2020 on the left, compared with images 2 years earlier (right)
Maralbeshi Facility (#6)
Left, the camp in 2019, very right, in 2016. I am unsure whether these are cotton fields, but the white/green colour could fit the description of cotton fields as portrayed below

Sports grounds: (basketball and other sports courts)

(Timeline: left to right) Recreational sports grounds built and removed again (this has nothing to do with Covid-19, courts got removed before the pandemic hit) — why isn't clear


10 suspected crematories in places where re-educational/detention facilities are dense
Three of the suspected crematories are in a 15km radius of facilities near/around Hotan
These are the three confirmed locations of crematories
Extracting information and exposing the government via the job ads is useful. Erkin Sidick, president of Uyghur Projects Foundation posted on twitter in Sept. what looks like a job ad to sell forced labour (I couldn't verify it, but others, like Timothy Grose (@GroseTimothy) corroborated the allegation by posting evidence for similar job ads posted on Baidu, with others posting about it too. I strongly stress, these need to be verified)

What population/urbanisation numbers tell us about internment

What can exports tell us about forced labour?

Right: Alataw pass railway station (and left, where it is on the larger map, last stop before exports touch ground in Kazakhstan)

Baidu maps: Checking what the Chinese tech companies are ‘hiding’:

2019 (Google Earth, CNES, Airbus) — and right, Baidu Maps

What can the camps in Tibet tell us about the camps in Xinjiang?

2018, 2019 and 2020 evolution of the suspected Botuocun camp in Tibet (31.0958 97.2035) The camps dissolved and in 2020 there was no trace of it anymore. Why was is removed?
The second camp in Tibet (31.5378 92.0250) — heavily fenced, a place cars are parked and lots of blue-roofed factory buildings, which could house forced labour
Some details on the added structure for the one above.
Third Camp (29.3177 88.8146): High-security prison features, tall walls, with a dedicated prison wing and connections between them

Closing remarks:

Investigative journalist with a technical edge, interested in open source investigations, satellite imgs, R, python, AI, data journalism and injustice

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store