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 pressing need to investigate characteristics of Xinjiang’s detention camps
The story has been widely covered. Calls by human rights advocates to define China’s practices as ‘genocide’ grow louder. Hundreds of thousands of Uighurs, Kazakhs and other Muslims detained in internment camps. Many still are.
“Inmates undergo months or years of indoctrination and interrogation aimed at transforming them into secular and loyal supporters of the party”, the New York Times wrote and published documents that unmistakably prove a dire human rights situation in the west of China.
First China denied the camps ever existed. Then the Chinese consulate doesn’t bother anymore to play a smoke and mirror game and admits: “Xinjiang has set up vocational education and training centres in order to root out extreme thoughts…”. Their purpose: ‘compulsory programs for terrorist criminals’.
Now, the language changed again. China’s President said the ‘strategy for governing Xinjiang in the new era is completely correct.’
Unacceptable (and unwise) of some to deny it. Social media commentators, some who are frequently quoted by large media organisations, keep casting doubt on the tragic story. Margaret_Kimberley tweeted — after an ITV news report emerged — “These are lies. There is no evidence of Uighur concentration camps. More hybrid war against China” (it received 2,000 likes).
While there is no room left to doubt that these camps do exist, there remains vast uncertainty whether investigative journalists and human rights advocates located all facilities spread out across the province.
Researchers/journalists who made it their beat to find them, like Nathan Ruser at Australian Strategic Policy Institute (ASPI), says “we don’t believe that we have found them all”, after posting 380 facilities online.
Germany’s chancellor last week said China’s President Xi offered delegates to send envoys to visit Xinjiang province [and camps] to see for themselves. Chances increase to see more of the so-called ‘show camps’ for a short period of time or as long as the visits take (the BBC encountered it when it visited last time). Xi also ensured that there will be an ‘ongoing human-rights dialogue’. But Ursula von der Leyen tweeted “a lot remains to be done .. in other chapters of our relations”.
Satelite investigations exposing more and more evidence. OSINT journalists rely on them. It’s one reason why some open-source intelligence journalism colleagues keep hearing rumours that some of the camps may have moved underground (e.g. detention in under-surface facilities) to hide from the spying eyes and scrutiny of satellite data analysts (we don't have proof for this thesis but I encourage you to reach out if you have evidence).
Mounting number of facilities
The number of confirmed facilities steadily grew. A 2018 BBC investigation looked at 101 campsites, which got pinpointed via various media reports and academic research, the author says.
Most recently, Buzzfeed investigated 268 compounds, many from previous lists I worked on too. In February, the list of ‘confirmed re-education camps’, so lower-security sites, mainly for indoctrination purposes, was limited to mere 50 facilities. ‘Confirmed’ in this context means they have been validated by eye-witness reports. Back then, there were another 170 that had yet to be confirmed.
It is of vital importance to keep this investigation rolling. This means to forensically document the changes in these camps and to spend more time on characterizing each detail. ASPI just dropped a new list and we are going to work with that one instead of the original 50 we received (the list can be downloaded here and geodata that can be simply dragged and dropped into QGIS and Google Earth Pro, it is available here).
Finally, news broke via Reuters (and research by Adrian Zenz) that evidence of forced labour is mounting also in Tibet (we will look into this later, too).
List of ‘expanded camps’ extended
Earlier in the year US-based Uighure group ETNAM shared a list with around 50 confirmed sites. We and others scrutinised this list on increased activity on the ground via aggregated satellite remote sensing data (link). The list was shared as klm. file. It helped enormously with going through them one by one. All the coordinates as well as the Chinese names of the places are accessible via Google Earth Pro. Now that ASPI dropped a new list with coordinates and updated 2020 records, some of the work we have started can be extended and match.
Because we are most interested in the camps that got expanded (so buildings or features were added), we will concentrate on the list of facilities that were developed. It includes a list of 61 sites.
Why is the onus on expanded camps? In addition to the characteristics ASPI added as classifiers, the extended camps might tell us where the local administration invests and where forced labour in the firm of Uighur prisoners went. We added a few more details for each facility that we thought was worth looking at (see sheet above).
We will go through various ways to characterise/investigate facilities and their surroundings
First significant markers includes the size of the camps. That includes quantitative details such as the number of buildings on the premise and adjacent to it. We will go through how to compare them. There are the walls of camps that are usually quite straight-lined. Their height, which we will define and validate, and the walls’ thickness may tell us something about recent developments (e.g. how secure the sites are, or were meant to be).
Guard towners are also a quantifiable element. ASPI and others counted them. Because they can be seen from outside they may act as a signal to local residents. That is also likely the reason why those facilities that have some or all of their towers removed recently tend to locate closer to residential buildings (see my stats below).
These changes are further revealing as they may tell us something about how the local government in various parts of the region varied in their response to international pressure (or not, by keeping them in place). ‘A lot [camps] had their security features removed in the second half of 2019’, Zenz explained. Some remained in place (important to add here, it remains doubtful that conditions improved inside of the camps, even if towers or security features were removed).
Zenz has an explanation for some of the changes: “On the same time they invited all these delegations and visitors, they released a lot of people. If you release a lot of people, you can afford to run with fewer security features. That can still be run like an internment camp, I’m sure”. We will look closer at what has changed ourselves.
Including those features above, there are a number of other aspects to take into account. We put them into the list below — each will be discussed separately:
- What blue factory buildings in and around camps can tell us
- What typical ‘prison features’ tell us
- What cars in parking lots tell us about personnel working at the facilities during Covid-19
- What walls can tell us
- What guard towers can tell us
- What sports facilities can tell us
- What the shapes/types of buildings and location can tell us
- What agricultural space (e.g. fields) around the camps can tell us
- What potential crematory sites reveal
- What Xinjiang’s export tell us
- What population/urbanisation numbers tell us about internment and surveillance
- What Baidu maps can tell us
Blue-roofed factory buildings
In satellite images, they are very pronounced with their blue coating. They may also heat up in the summer.
Most of them are factory buildings, has been reported. You can see them added in and around camp facilities, whether they are low or high security premises.
We can quantify them by counting them or via quantifying the space they take up. ASPIT decided to count them, though some buildings are smaller and other are massive. Google Earth has a polygon area measuring tool. A third option is to write a statistical model to calculate square meters factory floor space. If you are lazy you can consult a service that helps you with that via a visual detection algorithm — it calculates the area and records the number of blue roof buildings for a given satellite image.
One of the camps that expanded in the past two years is the tier 1 low-security re-education facility in Bugur in Bayingolu (41.808855284.3005783). It has a dense network of factory buildings nearby (around 23) and within its own walls there are eight. We used ASPI’s data to confirm this that noted: ‘considerable room for expansion’.
Let’s run the classification system over it and classify how much blue-roofed buildings that scatter around the camp can we count (importantly not all are factory spaces but many will be).
On the AI model: I downloaded the images with their highest resolution from Google Earth. To make the image a bit clearer for the model, I adjusted the brightness, upped the contrast and tinkered with the exposure. We can see the blue buildings, roughly in a radius of 1.5 to 2 miles (see image), account for about 1,464.9 m² (0.15ha). The number of little blue buildings expanded considerably since 2014 where they accounted for 1,022m2 (0.10 ha) — sadly we only have an image for 2014 and one for 2019.
Short intersection on the availability of images available in Google Earth:
Some of the important images to document the progression of these camps are missing. Some camps have a mere handful of publically available images (as in the case above). This is appalling and private satellite image companies need to be nudged to make more images public. Especially for the latest developments, this is urgently needed. Researchers noted down the latest dates for which images are available at the time of writing. Below we see them grouped by months, and then by facility category (tier 1 to 4).
What about bias to provide fewer updates on higher-security facilities? We don't have much to go in here (there is no direct evidence that western satellite companies are being pressured into not publishing their images for camps on Google). Despite only a few camps that didn't get updated at all over the past two years, we can see at the time of writing that Google and others hold more images for lower tier facilities (1 and 2) than for higher-security facilities (tier 3 and 4):
Continuing on the factories, another example is the facility in Maralbeshi County (39°49'7.84"N, 78°31'4.37"E). It was erected around 2017/2018. In Google Earth, you can see how the blue-roofed buildings surround the internment complex. Note, how the larger blue factory complexes to the left and right were there before the camp was erected.
In other words, the camp was planned and embedded into existing factory operation. It further corroborates a thesis that factory work by prisoners (in the form of forced labour), was part of a grander plan all along (though, to be certain, looking at satellite images alone does not suffice).
Adrain Zenz thinks blue roof factories is something that warrants looking into in more detail. A bunch of these blue roof factory building were erected in 2018, especially in the second half. Zenz explains it’s important timing because the policy documents on forced labour, as explained in his post from last December, shows that a lot of this kind of policy was released in the first half or mid of 2018.
A recent Buzzfeed investigation did mention blue roofs but surprisingly didn't pay more attention to the matter. The factories grow in importance as the forced labour of imprisoned groups is being increasingly ‘commercialised’.
ASPI’s data recorded the distance (measured in km I assume) between the 380 facilitates and the local/nearest industrial parks — where some of the forced labour could have moved to put to work. The data categorizes facilities in four areas of security (ranging from Tier 1= re-education camp to Tier 4= prison facility). Tear two and tier three camps tend to be located more closely to the industrial centre of the towns, the data suggests (see chart below):
Zenz adds: “what’s significant is the sudden increase of blue roof, single story, flat type factory buildings. It’s consistent with policy, and also release, the Karakax list also talks about people being released into forced labour. A lot of that took place in 2019.”
The blue metal barracks found in Dabancheng shining light yellow in the sentinel IR images as they are being reflected. Low res Sentinel 2 data also suggests that these metal-like structures in the south of the Payzawat camp (Payzawat County, 39.538372, 76.713606) may also heat up in the summer. SWIR (short-wave infrared imagery) and NIR can be used for heat monitoring.
Prisons features: camps that imprisoned people become more ‘secure’ not less:
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.
While it is true that some camps removed some of the towers and other security features (labelled ‘desecuritisation’ by ASPI’s records), others increased theirs. Those happened to be facilities that are detention centres and prison. In the context that Chinese authorities moved prisoners to these more secure facilities with less transparency and harsher treatments, this is cause for concern.
Let’s look at an example. From the list of expanded camps, there is the camps Yarkant Facility in the Kashgar prefecture (38.351531177.3055467). Since 2018, we saw a nearly 10,000 m2 large factory compound built (compare images from 5/8/2018 with 1/21/2018). Then, a year later, watch downers got added. There are now 8 towners. For such a small facility that’s quite conspicuous. The reason it’s a high-security prison facility.
Newly built detention/prison facilities created between 2018 and 2020 are of special interest. Camps like the tier 3 (detention) camp of Sanji Facility (#3, 44.102764,86.9960751), a with several watchtowers and an external wall is important as we can follow the progression of each step of the building process with high-resolution images.
The location was probably chosen because of a lower-security area nearby, north of the facility (3/7/2018). Building must have started in the summer. A couple of months after the last shot (8/11/2018) the blue-roofed factory gets built-in the north-west of the camp (a reason to assume a direct relationship there) and within two weeks in August the main building takes shape. At the same time, the walls get erected and we can make out the layout of the facility with its heavy concrete structures.
We can see, those are fundamentally different from building built in other lower-security camps. Then two months later it’s almost completed.
The speed of building is noteworthy (better trackable if we had access to a more continuous stream of images). From the few images we have above and those from Sentinel 2, below, we can assume that it took the developers between three to four months in pure building time to pull it up — an astonishing pace. China is renowned for its fast building pace. For many other areas, such as coal plants and artificial island-building its cookie-cutter approach — where blueprints are being re-used over and over again - it permits building more quickly.
Other who looked at the situation in Xinjiang reported that many Uighurs held in lower-tier facilities could have been moved/transferred to higher-tier prisons. In other words, despite some re-education camps have experienced ‘de-securitisation’, half of the camps that expanded are higher security facilities, so tier 3 (detention) or tier 4 (prison) camp facilities.
What parking lots tell us about the camps during Covid-19
I believe this topic has largely remained unexplored. Busy parking lots are one way to tell how many staff members are on site. Especially interesting it this for the recent month that were affected by coronavirus. We dont know much about the conditions inside of the facilities.
But with fewer staff members around (and fewer visitors allowed — previous reporting has revealed that detention centres have ‘small visitor centres’), the lives of inmates may have worsened. There was some reporting that Covid-19 cases spiralled in the province of Xinjiang and some expressed concern that cases could spread within camps. It’s possible, no doubt. With only a few cases in the whole region, though, the risk is lower.
Pandemic related fears may have affected the material and food supply. Sick imprisoned detainees may go without healthcare treatment for weeks or months. All these are assumptions for which we have little evidence. But the possibility alone raises concerns. If it is true that prisoners remained in the facilities during Covid, they could have suffered from the absence of staff and proper care.
From satellite images, it is hard to know — though there is some evidence from an eyewitness account shared by a historian, a Georgetown professor on his Medium page.
We might be able to tell how many temporary people were on sites (those that use their car to leave for the night). Counting vehicles at nearby car parks is one way.
At some facilities, we can clearly see the parking lot. An example is Ghulja City (43°58'37.52"N, 81° 8'18.98"E). It’s a fairly large car park. We can use Picterra system (there is a 10 day free trial version) to check the satellite images for May 23 — thought there isn't much to count, the car park is empty.
Seven months earlier, on October 24th of 2019, we count around 120 cars (with some false positives, but that’s good enough for us). The algo gives you a count so you don’t have to count the red boxes one by one. Once trained, we can run it on subsequent images.
Let’s walk you through how to train and count the cars. I simplify here (a more complete tutorial can be found here and in their platform). First, we use one of the images to train the algorithm on the cars in the car park. Then we run it on the other pictures. It’s neat and simple (and quick if you don't have time to run your own statistical model in python).
The number of vehicles dropped during the heights of Covid-19.
We could do this for other confirmed location such as the facility in Chochek City (Tǎchéng Shì, 46°43'3.79"N, 82°57'15.23"E) where car numbers dropped in April. We see this in many other facilities (for those that expanded).
Hotan City Facility #1 (37.1117019, 79.9711546) with 81 cars in the parking lot at the end of 2019 dropped to 10 during the height of the pandemic. Similar developments have been perceived at Hotan County Facility 1 (37.2420734 79.8595074), Ghulja Facility 1 (43.9756437 81.5009539) and a number of others.
Calculating rooms and capacity
How many people fit in a facility. If we take the example of the re-education camp in Chochek City ( 46°43'3.79"N, 82°57'15.23"E), we have high res Google images for the end of March and end of April of 2020. We can see the thin middle part is three stories high and in earlier images (Jul 18, 19) we can see the southern part is four stories high. In 2018, we got an image of the foundation when it was built. This provides enough detail to calculate that the facility has around 367 rooms — for the total t-shaped building with the arms.
In the example above, we shouldn't be too sure that alls detainees were kept in the facility during Covid. Some reports claim that some of the other lower security re-education centres kept people ‘only during the day for indoctrination classes’ (it’s certainly different for the high-security prison facility that is also on the premise of the Payzawat facility, see in the south, with their towers).
Comparing camp sizes
The total size of the camps matters, especially when they get extended. Most of the camps have clear wall frames build around them. It’s one of the most important and simple characteristics. The wall frames makes it relatively easy to draw shapes in your geolocation system of choice (the sheer size of the walls, might be less ideal to gauge the number of prisoners).
Some have vast empty space in between might suggest that other faculty sections or factory buildings are due to be added. Some are cramped with building.
Tracing and calculating the area of wall frames in Google Earth for some of the largest camps, we get what we already knew:
To emulate the work ASPI’s data was posted here. A number of track and trace tutorials for Google Earth (one here on measuring property space) are available on YouTube.
Staking out camp size:
The Qariqash County/قاراقاش ناھىيىسى /墨玉县(Mòyù Xiàn, 37° 6'44.88"N, 79°38'32.71"E) sits in the South of the large stretch of desert.
We use the polygon tool in Google Earth to stake out the clearly marked walls. You usually end up with a rectangle. Under measurements (right-click on the item) you can see the perimeter is around 1.65 km and the area is roughly 16.7 hectares (0.17 square km).
Now we can compare it with another one on the list, the camp in Aqsu City (41°11'27.12"N, 80°16'25.08"E). It’s markedly smaller, with a perimeter of 1.1km and only an area of 5.65 hectares. There are other ways to do this in QGIS, a geoinformation system more efficiently.
What can walls and towers tell us?
How tall are walls at some of the camps? The answer varies across the vast variety of facilities. Height may tell us something about who built the camp and the level of security. It’s unsurprising to find different heights at different camps built by different planners.
Where we don't have shades available, we can check the two images above and reference them with the people in the image and define the height this way. Another standard way to calculate height is using the shades by the walls and towers and calculate the height via Google Earth and SunCalc.
The shade of the southern wall in the satellite image from 03/19/2020 for the Dabancheng camp is around 7.62 meters long. The towers on the southern wall for those dates result in a height of around ~8meters.
But the images in the Reuters shots look different. That’s why they were taken a year or two earlier. Satellite images from 4/22/2018 show clearly the octagonal shapes of the tower shades. If we calculate again, the shade of the tower is around 9 meters long, translating into around 14 meters in height.
We do this for the wall as well. What we find is that, although the towers disappeared (though, some are still there, just not protruding so visibly), the only thing that really changed is the height of the walls — now around 13.5m tall, compared with 9.5m in 2018. The same towners, removed from one Dabancheng camp, then re-emerged half a kilometre south-east at the other newly built one (2019).
Why are we even bothering measuring height? On one hand we want to answer how security changed across the camps. Are walls getting higher? Do they change in their layout. It helps to classify the type of camps. The higher the walls, the more secure they probably were meant to be. Higher wall might mean higher chance that prisoners are held at facilities over night. It also may help to disprove claims by XJ denialists.
We can verify the Suncalc analysis with images. Cherchen County, for which we reviewed images for 12/14/19 shows roughly the same height. Explainer how to measure the height of an object from satellite image available here and here.
The number of press images of the camps is limited. Most are by Reuters or AFP/Badung Police. It is this one here (37°14'29.78"N, 79°51'35.00"E). More local street footage, though not of camps, might be obtainable via Mapillary.
Buildings shapes/outlines and location of camps
Let’s start with the location of the facilities first. ASPI recorded the type of security for its 380 odd facilities, and for many the distance to populated areas such as residential buildings. When local administration planned on where to place the facilities they might have taken into account how the neighbouring public should (or shouldn't) perceived them. More secluded camps are more hidden from public scrutiny. Those near people’s homes or schools may be placed there to have the opposite effect.
What’s immediately apparent when running a few inferential statistics on the records is that the more secure detention centres tend to be kept further away from buzzing residential areas — meaning, further away than for instance Tier 1 re-education camps, which are often nestled between residential parts of cities, or occupying old schools.
Agriculture/fields around the camps — investigating forced labour by detainees
Identifying agricultural fields near or around facilities may reveal some potential aspects of how forced labour in the camps were used in close vicinity.
Especially for secluded faculties, with not much else urban life going on (so reducing the possibility that other local farmers were involved in working the them), the chance increases that Uighurs detained were used.
One example is the facility near Yingye’ercun, in Gulja, with a 0.16km2 large campground (43°58'37.52"N, 81° 8'18.98"E). The farming area that was developed since 2018 (shortly after the multistorey buildings was built in the core of the facility) spans 1.7km2 and is clearly marked (which includes the facility itself, see in red below).
In other words, once the camp was built the fields surrounding it got worked and developed— unlikely to be only a convenient coincident. The nearby factory complex was also extended.
Often it warrant also checking with Sentinel 2 images on EO browser. In this case, it’s useful because it allows us to visualise agricultural development via its invisible light remote sensing capabilities. Additional bands (which Google images lack) give access to the invisible spectrum and shows the agricultural expansion (here shown in red via the false colour composite, commonly used to assess plant density and health, “since plants reflect near-infrared and green light, while they absorb red”. Exposed ground are grey or tan, vegetation is red).
Another camp in this regard is the Maralbeshi Facility (#6) in Kashgar (39.7406222 78.0115086) with lots of fields surrounding it.
Why is the forced labour aspect in Xinjiang’s agriculture so important in this debate? For one, it’s part of the human rights abuse that more and more governments and industry leaders recognise (such as Swedish company H&M, who profited from cotton supplies and other kinds within their supply chain). Some decided to cut ties with suppliers in the region. It may the answer for the short term. In the long run, western businesses much apply pressure to get suppliers on their own to dissuade local forced labour practices (see example on ads that emerged to sell Uighur forced labour online).
According to the ILO Forced Labour Convention from 1930, forced or compulsory labour is defined as ‘all work or service which is exacted from any person under the threat of a penalty and for which the person has not offered himself or herself voluntarily’.
Sports grounds: (basketball and other sports courts)
Some found value in observing their development. BBC’s John Sudworth found that just before a press tour organised for his press teams two years ago the appearance of recreational areas altered. In some of the places they were taken to, satellite images and the internal security fencing — and what looked like watchtowers- where taken down shortly before the tours for journalists began. Specifically on sports grounds, they noticed that empty exercise yards have been transformed into sports facilities.
The reporters asked: if the journalists have been presented with mere ‘show camps’, what may this say about the places they were not taken to. Sport facilities are quite easy to spot from satellite. The BBC travelled to Kaxgar in the very east of the region, about 100km south of Kazakhstan’s border. Their footages shows how the camp put up courts shortly before the press trip. But they didn't last long. We found evidence that these very courts disappeared again in early 2020 (see below).
In one of the camps in Qariqash (37°15'32.54"N, 79°44'52.08"E) the sports facilities were made unavailable as recently as July. Now big brown sheets, what looks like blankets with knobs on them, cover them. Those have never appeared on satellite images before and extend to the soccer field in the north and the big parking lot next to the sports courts.
I have mixed feelings about recreational activities. We must strongly doubt that they benefit people held for indoctrination. So are they only a smoke and mirror game to show the friendliness of re-educational camps? Or are they actually benefiting the imprisoned? It is hard to say. In recent time, they are more likely to be added than removed. In around 37 facilities on the ASPI list basketball courts, running tracks or other sports fields were noted to have been added or extended.
When we compare the average distance of residential building for these places (1.2km) with the average distance of all the places where we have a record on the distance to buildings (1.8km), we find the recreational activities might be used as an element to signal the locals that the facilities have those recreational features.
Dabancheng has one court in the western block and a number of other ones in the centre part. In the eastern wing, there is nothing. We haven't got any further high res satellite images on Dabancheng (other than those until March 2020, that leaves only checking Sentinel 2 images or commercial images).
I am going to stop here. The analysis of recreational areas yielded rather little, for me and the folks at ASPI. “I don’t think the sports grounds mean much in the detention regime”, Nathan Ruser says. If you have more info do reach out or leave a comment.
Crematories
The New York Times followed the lead of findings (that emerged last year, also mentioned in the state.gov report) and check the extent of description of religious sites and burial grounds. In September, the team reported that ‘thousands of religious sites’, such as mosques, shrines and other sites were bulldozed or replaced.
As many burial grounds disappeared and people within camps families have never heard from again, the question of how Uighurs’ life proceeded became more pressing. Crematories may be one aspect. Some anecdotal evidence by a source spoke of a nascent growth of crematory sites in the areas near camps. This appears important in the context of how prisoners are treated in facilities and what happens if they die and at what rates.
High prevalence of tuberculosis in facilities worries insiders. TB is spread via droplets through the air by someone who is infected. It’s especially deadly when the immune system of those who caught it, can’t cope with it. With the conditions reported by some of the eyewitnesses, it is feasible that the hard conditions prisoners are being subjected to, could enhance the deadliness of TB.
The think tank which produced a previous list of facilities searched and found a handful of crematories (I don't think they concluded the research and it continues, perhaps with your help of OSINT research).
The reason why crematories are of interest is that Uighur are Muslim, Muslims don’t burn the bodies of their dead. They bury them (creation is strictly forbidden). Seeing more crematories pop up might be a first clue on whether dead bodies from detention facilities are being burned. We have to stress here, we have to be extremely careful with drawing quick conclusions, the base of evidence is thin. One would need to check local statistics and cross-examine them with other data source.
We will concentrate only on the sites itself. The ‘unconfirmed sample of crematory’ consists of ten sites. These are listed below. Just a word of warning. Feel free to investigate them further — either via additional satellite footage or on-site visits. Nonetheless, these get us started. The first three are confirmed by eyewitness accounts or local records (as far as I was told, this is sadly only secondary research).
Cr_Gholja_01 (Existed, 44° 0'17.86"N, 81°13'40.43"E); Cr_Artush_01(Existed, 39°44'35.47"N, 76°12'7.49"E); Urumchi 2 Funeral Parlor (Existed, 43°54'55.20"N; 87°36'9.01"E)
Cri_Hotan01_(Suspected)
Cr_Artush_02 (Suspected)
Cr_Hotan_02(Suspected)
Cr_Urumqi_02 (Suspected)
Cr_Urumqi_01 (Suspected)
Cr_Urumqi_01(Suspected)
CrArtush_02 (Suspected)
Now let’s take a look at the characteristics of the confirmed crematories. They have some distinctive shapes, including a rectangular architecture, walls or a treeline that fence the premises (framed in black). Where marked ‘burial grounds’, I was unable to confirm this but checked with a few other sites mentioned in the coverage that was exposed in 2019 and it looked similar (in short, more time needs to be spent on this).
What helped the researchers identify the confirmed ones? According to the source, the Chinese called them ‘burial management facilities’. It’s apparently a euphuism for ‘crematories’. The Chinese government bulldozed some burial grounds with the justification that they would take up too much space which was covered in the 2019 reporting.
The other aspect is whether relatives receive the body of loved ones that die in the camps. Salih Hudayar (now Prime Minister of the East Turkistan Government-in-Exile) says he had a relative who died in a facility (he don’t know whether in the camps or the prison) and his family was not able to have his body returned. He thinks that many other Uighurs have not had the body of a deceased family members returned to them. He assumes they are being cremated as no record exists of a burial site.
More crematories are only possible if you have employees who staff and run them. The Chinese government tried to find those employees online. “We assume they are being cremated because the government ran job ads and offering high salaries to work on these [crematory] sites”, he added.
The suspected crematory facilities were then modelled upon the layout of the existing/confirmed ones — e.g. compared with buildings in and around the area. “We found a couple, but we are not 100% sure”, the source admits. Here OSINT journalists could become useful (let me know if you have intel on this matter to follow up with).
On the description in 2019: evidence surfaced that 45 Uighur cemeteries have been destroyed since 2014, including 30 in just the past two years (research was carried out by AFP and satellite imagery by Earthrise Alliance, here reported by the SCMP).
What population/urbanisation numbers tell us about internment
Salih Hudayar explained that what worries him is that population statistics don't square. An often-cited figure of 7 million Uighurs in the province is much lower than the official estimates of the Uighur people.
The number often used is 12 million Turkic-speaking Muslim Uighurs. The number could be higher. Especially in the villages — Uighurs are allowed to have only three kids — some families have more than that and don't register their offspring, as a result, many kids lack birth certificates. Other figures on the number of Uighur population is much taller (larger than twice of the 12 million figure, but remains hard to confirm that. The closes figure the Chinese government will have internally after the government’s extensive and invasive security and surveillance campaigns, in part to gain information regarding individuals’ religious adherence and practices).
The rising number of orphanages and kindergartens is also of interest. A satellite and local administrative data analysis should track them. The premise here: the more aggressive the detention of families are in XJ (moving Uighurs from low to higher security facilities), demand for places that house children increases. More orphanages and child-caring facilities could be revealed.
What can exports tell us about forced labour?
The type of exports of a region can help to figures out what to look for when it comes to forced labour. Increasingly, the international textile and fashion industry wakes up to reputational damage if supply chains incorporate Xinjiang forced labour. EU leaders held a meeting with China’s president Xi last week where Xi ‘rejected’ foreign [political] meddling in his nation’s affairs. But businesses have more leverage. Xinjiang is busy trading with foreign powers. The Chinese province accounted for a large part of the world’s supply in cotton. Exports amounted to $19.3bn according to export documents (export data for the west of China can be found in China’s official data stats, Stats.gov.cn, customs.gov.cn, or mofcom — this might be useful. Comparing what the government reports and what’s happening on the ground might reveal discrepancies, as it did before).
Exports (to Europe, across the silk road to the west) is directly connected at A busy train station connecting to the neighbouring country of Kazakstan in the northeast (the export route is called Ala Pass. A short promotional video here). Given the rebound of the Chinese economy, the shipments/trainloads must have increased in May after the effects of the pandemic subsided. What’s unclear is to what extent and whether that matches what the government said.
Satellite images might reveal discrepancies when train containers at the Dzungarian Gate (the Dzungarian Alatau mountain range along the border between Kazakhstan and the Xinjiang Uyghur Autonomous Region) are analysed. It’s the main connection between China and the west.
The main railway station in Xinjiang for the Alataw pass is the Alashankou railway station (situated here: 45°10′13″N 82°34′13″E). It’s the last resort for export containers before entering Kazakhstan.
OSINT journalists may be able to gauge Xinjiang export traffic by counting the number of containers on rail tracks. It might be laborious effort, not sure if it yields anything.
More useful would it be to monitor the use of agriculture and factories in the nearby vicinity of camps, as shown before. Or perhaps they can be linked up.
Baidu maps: Checking what the Chinese tech companies are ‘hiding’:
The Chinese government may have little interest to showcase their human-rights violations which they deem as justified (Xi’s statement). Satellite images on Baidu Maps show maps that hide most of the facility. What to make of it? Google Earth lets you upload so-called ‘overlays’. If you stretch them to the right size you can compare the uploaded screenshot (we took from Baidu) with those present in Google Earth. For Tumshuq City/تۇمشۇق شەھىرى/图木舒克市(Túmùshūkè Shì) (39°54'40.02"N, 79° 1'26.09"E), see below.
Why is Baidu’s involvement increasing relevant? On one hand, it is important to see the connection between private sector companies and the government. Chinese satellites are able to update and provide high-resolution images to the maps on Baidu. But they don't. We had a similar debate on Twitter, that some government used to press companies to blur our images. But because images are available on other platforms ‘unblurred’, the practice was largely discontinued (there are still examples but they are getting fewer). One reason is that if a blurred area appears, it signals others to be extra vigilant and look out for other images. Instead, what increasing happens is that companies with private satellite are ordered not to release them (read more about the debate here).
Baidu map’s decision to not show images on certain facilities have backfired. It can be reverse-engineered. Areas where images are unavailable became extra interesting. In this way Buzzfeed used Baidu Maps to their advantage. They located/confirmed some of the camps because of it. This way, they turned shortcoming into an opportunity. You may want to be quick in replicating this principle for other parts of the country where forced labour/detention camps are expected (e.g. Tibet). Such loopholes will usually be fixed swiftly.
Bit more on the tech. According to a 2019 report by Human Rights Watch, Baidu’s map function used in the IJOP app, a controversial system used by the police and the state that generates “a massive dataset of personal information, and of police behaviour and movements in Xinjiang (it is not known how the authorities plan to use such data): The IJOP app logs the police officer’s GPS locations and other identifying information when they submit information to the IJOP app. The IJOP app uses a map functionality by Baidu, a major Chinese technology company, for purposes including planning the shortest route for police vehicle and officers on foot, according to the app’s source code.
What can the camps in Tibet tell us about the camps in Xinjiang?
Reuters reported just last week that forced labour expanded to Tibet (south of XJ). Reuter’s own reporting corroborated the findings obtained by Adrian Zenz. It would take another post to go into how to investigate the state of transferred Tibetan labourers. The quick and dirty check on the situation shows the merit of using satellite images to investigate grows as foreign journalists are being barred from areas, such as entering the Tibet region (foreign citizens are only permitted on government-approved tours). OSINT lessons from investigating XJ should be applied to Tibet too.
How does Xinjiang link to Tibet? The former Tibet Communist Party Secretary Chen Quanguo was chosen for the same job in Xinjiang in 2016 and headed the development of Xinjiang’s camp system, Reuters reported.
Mass incarceration started before Quanguo came onto the scene: A fanghuiju work team was dispatched to a village in Guma wherein 38 individuals were allegedly detained in a government campaign, in early 2016 — it’s true however that Party Secretary Quanguo, appointed in August 2016, who waged a ‘Strike Hard Campaign’ against violent activities and terrorism increased repression.
In an article last year, The Print used satellite images to prove that at least three Tibetan “re-education camps” are currently under construction. The author of the survey was Vinayak Bha, an ex-colonel retired from the Indian military intelligence unit.
Col Vinayak Bhat (@rajfortyseven on Twitter) found three camps in 2018/2019 and share them. One of them is the one in Botuocun (see below). Bha writes about Chinese military deployment dynamics. The temple of Tibetan Buddhism is a ‘concentration camp’ that is surrounded by high walls and guard towers and has the same structural design as a prison. It is feasible that China’s mass detention to spread to Tibetans. Methods will likely base on the model executed in XJ.
The reports of the three camps emerged in 2019. “Small-scale versions of similar military-style training initiatives have existed in the region for over a decade, but construction of new facilities increased sharply in 2016, and recent policy documents call for more investment in such sites”, one report stated. Looking at the three sites, some of them are quite old but the one below is less than three years old.
The allegation is that these facilities are now be used as detention centres for political indoctrination. “The detainees are allegedly used as forced labour in government factories and projects during the day time or as per shift timings”. It is something that rings true under the light of camps in Xinjiang but we lack evidence from the satellite images.
There is some evidence that additional factory buildings were added. For the facility above, buildings in the upper east wing, with red roofing was added recently. Their layout reminds us of the blue-roofed buildings in and scattered around Xinjiang facilities, which we also have present: “This architecture is bang on a XJ prison, [though] with a different style roof”, Ruser said.
The prison layout from the older prison facility above — with its long and vertically arranged wings and the rippled features — is similar to prisons seen in Xinjian, such as the two portrayed below (one at Qariqash County at 37° 6'44.88"N, 79°38'32.71"E and the other facility in 39°25'54.60”N, 76° 3'20.59"E).
Closing remarks:
There is a mountain of stuff not included here. This is a training post and not an investigation with full-rested conclusion. This post should encourage other open-source investigative journalists to look into the facilities, follow their own reporting and help monitor developments/details that others may have missed.
At present there are only a handful of OSINT journalists looking into it. Even fewer have the time to continuously keep this rolling, e.g. analysing the camps as other stories press them to move on.
We need more eyes on this. The alleged human right abuse must receive all the international scrutiny it can get. People like Shawn Zhang and others with Nathan Ruser and APSI) started the journey. Other journalists must continue and expand on it.
Also, the more open we are about sources and the analysis (hopefully) the fewer people might try to cast doubt on the existence of the camps (good thread here)
OSINT techniques used must master the skill to help others to replicate the findings, step by step. That’s the reason this post resulted more in a hands-on tutorial than an explanatory post. I encourage anyone to start looking into the human rights abuse (though, I must stress, be careful to draw quick conclusions. Instead, share what you see on satellite images with the community of serious journalists and OSINT investigators).
One last thought on commercial satellite imagery companies. It is crucial to get their support on this. For more than 100 camps mentioned in the latest update of the ASPI list (nearly 80 of them high-security detention facilities — classified as tier 3 or 4), we have no updated record of satellite images. This leaves researchers and journalists only to low-resolution devices, by Sentinel 2 images, or beg for images from Maxar or Planet Labs. That’s not good enough. Transparency requires companies inc to make those high-resolution images available, to anyone. Intelligence services should also consider making their high-resolution images available to the public for scrutiny, though, that unlikely to happen.
