A growing body of research has systematically documented Chinese efforts to detain, re-educate, and imprison its ethnic Uyghur and minority groups throughout western Xinjiang Province. Existing estimates suggest that between 1 million and 1.5 million Uyghurs and other minority groups may be detained in such facilities, subject to alleged reeducation and indoctrination, forced labor, false imprisonment, and even claims of physical and sexual abuse.

As international pressure condemning these practices has grown in recent years, Chinese officials have variously denied the existence of these facilities, suggested that they serve only as benign vocational training centers, or that they are necessary to cure ideological diseases in those with problematic ideas sympathetic to radical causes.

By mid-2019, the Chinese government began to suggest that the majority of such facilities, whatever their purpose, were no longer operational, either because vocational trainees had simply graduated and now settled into stable employment, or that released detainees have freed themselves from the influence of terrorism and religious extremism. Yet a steady flow of research and original reporting has since cast doubt on these claims, employing a mix of on-the-ground reporting from existing facilities, official documents, satellite imagery, and even nighttime lighting analysis to help identify new and growing facilities since 2019.

To expand upon this research, this three-part series will use granular monthly data on nighttime lighting in Xinjiang to empirically assess the growth and decline of 380 known detention facilities across Xinjiang, as originally geolocated by the Australian Strategic Policy Institute (ASPI). The goal of RAND's research presented in this series is to provide an objective and systematic review of the status of Xinjiang's detention system, using a novel methodological approach.

This report, the first in our series, introduces RAND's approach to measuring monthly trends in nighttime lighting over Xinjiang detention facilities, and discusses our efforts to empirically identify sustained periods of expansion and decline across these facilities over time. Then, using this methodology, it presents new analysis of the explosive growth of Xinjiang's detention facilities over the last decade, offering a novel and unvarnished look into China's coordinated efforts to construct an elaborate system of reeducation, detention, and imprisonment of its Uyghur and ethnic minority population in Xinjiang.

Data on Known Detention Facilities in Xinjiang

To assess the growth and decline in nighttime lighting over Xinjiang's detention facilities, we begin with data on the location of known facilities in Xinjiang collected in a seminal report from the Australian Strategic Policy Institute (ASPI) published in late 2020. Using satellite imagery, nighttime lighting data, and primary and secondary sources, ASPI researchers geolocated 380 known or suspected facilities across Xinjiang, and assessed their level of security, overall purpose, the presence of factories for forced labor, and any visible evidence of construction, desecuritization, or potential decommissioning of these facilities over time.

To get a sense of the geographic scope of this detention network, Figure One below maps the locations of all 380 detention facilities in Xinjiang from ASPI's dataset. ASPI categorizes each facility into one of four distinct tiers, based primarily upon its imagery signature and the level of security seen at each facility. ASPI's Tier One facilities are low security reeducation centers (108 in total), which are often connected to factories and may allow detainees to return to their own homes on the weekend. ASPI's Tier Two facilities are also reeducation centers (94 in total), but with higher security, and whose purpose is to rehabilitate detainees short of a prison environment. 

The final two tiers of facilities comprise the judicial component of Xinjiang’s broader detention system. ASPI’s Tier Three facilities (72 in total) are characterized by secure housing facilities collocated with recreational facilities and administrative buildings on larger compounds. ASPI's initial research suggests these facilities are high-security detention facilities. However, based upon other sources of imagery analysis of Xinjiang’s camps, and published Chinese construction regulations for prisons, we assess that these facilities are actually long-term prisons.

ASPI’s Tier Four facilities (107 in total) are characterized by fortified secure cell blocks with significant layers of security, largely without access to recreational or educational facilities seen in lower security facilities. While ASPI's initial research suggests that these facilities are maximum security prisons, we  assess that these facilities instead serve as administrative or pre-trial detention centers. The lack of recreational, administrative, or factory facilities common to most of these locations is consistent with their role housing shorter-term detainees.

Figure One. Location of Known Detention Facilities in Xinjiang Province

RAND's Nighttime Lighting Methodology

To assess the growth of these detention facilities over time, and assess their current operating status, this research leverages geospatial data on nighttime lighting in Xinjiang as a proxy for the level of activity at each facility over time. Overall, nighttime lighting data provide an equilibrium measure of electricity consumption at specific locations on the map, for a specific period in time. For detention facilities, changes over time in nighttime lighting may offer evidence of large-scale construction projects, new or expanded buildings, or alternatively, the closure of previously-occupied facilities.

As such, we use monthly trends in nighttime lighting to help illuminate potential changes in the operating status of detention facilities. This approach allows us to capture changes in the intensity of activity at detention facilities that may not be visible in overhead satellite imagery, and at a minimum, helps us tip-and-cue our analysis to focus on facilities of interest. This approach follows that used in prior RAND research, where nighttime lighting data was used to assess changes in economic activity in cities captured by the Islamic State in Iraq and Syria.

For this research, we calculate nighttime lighting estimates using monthly cloud-free composite raster images from the National Oceanic and Atmospheric Administration's (NOAA) Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band, made available via the Earth Observation Group at the Colorado School of Mines.

To assess nighttime lighting trends over each detention facility, we first reproject each composite nighttime lighting image from its original resolution (15 arc-seconds, or approximately 350 meters at 40-degrees latitude) to a finer-scale 30-meter resolution using simple linear interpolation. This improves our ability to accurately measure nighttime lighting for smaller facilities and helps to reduce potential bias from adjacent infrastructure. From these images, we then calculate the average nighttime lighting inside 200-meter, 500-meter, and 1-kilometer radius buffers around the center of each facility. Then, we use an automated algorithm to select the most appropriate buffer size for each facility, based upon its overall size and its proximity to nearby industrial facilities, urban areas, and other detention facilities. This approach helps further minimize, although not eliminate, potential bias from the nighttime lighting of adjacent infrastructure with potentially different trends in nighttime lighting.

The resulting monthly trendlines for each detention facility therefore represent the average raw nighttime lighting value from the NOAA VIIRS sensor over each facility. But because raw VIIRS values themselves prove relatively noisy over time, we take two additional steps to produce representative trendlines capable of providing useful insights into the level of activity at individual detention facilities. First, we normalize all nighttime lighting measurements for each facility to its value in January 2018, improving our ability to make meaningful comparisons of the relative growth of facilities over time across facilities. Second, we smooth these normalized estimates using a ten month double moving average that allows us to identify sustained, uninterrupted periods of growth and decline in the signature of each facility, punctuated by clear inflection points. Of note, we define uninterrupted periods of growth or decline as any six month period where the smoothed nighttime lighting estimate consistently rises or falls.

To show this approach in action, Figure Two below offers satellite imagery and nighttime lighting data from an example reeducation center in Xinjiang's southern Hotan Prefecture. In the top panel of this figure, overhead imagery from November 2019 shows the 200-meter buffer used to calculate average nighttime lighting for this center, used to minimize potential bias of adjacent nighttime lighting from nearby industrial facilities and residential areas in close proximity.

Figure Two. Example Nighttime Lighting Trajectory of a Detention Facility in Hotan Prefecture, Xinjiang

In the bottom panel of Figure Two, the average monthly nighttime lighting of this facility is shown for each month from January 2014 through May 2020 (the last month in which data is available), shown in blue. These values are normalized to the level of nighttime lighting seen over this facility in January 2018, in order to better assess relative changes over time. This approach allows us to easily assess the magnitude of changes in nighttime lighting over time at this facility. For example, nighttime lighting over this reeducation center in January 2016 was roughly 80 percent lower than in January 2018.

The ten month moving average of this normalized nighttime lighting estimates are then shown in red, working to flatten out the somewhat noisy monthly variations in nighttime lighting and help isolate clear trends in the growth and decline of this facility. As shaded in gray, we assess that this reeducation center experienced a major period of steady growth beginning in September 2015 and ending in November 2018. This was followed by a sustained period of decline in nighttime lighting through the end of our dataset in May 2020, as shaded in yellow.

Correlating Nighttime Lighting Growth with Construction Activity

To contextualize the real-world implications of changes in nighttime lighting, we can use overhead satellite imagery to correlate these changes with visible indicators at known detention facilities in Xinjiang. In this report, we focus specifically on growth across the Xinjiang detention system; our second report will focus on decline.

More broadly, we see two major types of growth at the facilities in our dataset. In some cases, major growth in nighttime lighting at the location of a known detention facility occurs when that facility is built from scratch on formerly vacant land, often in rural areas with minimal existing infrastructure. In these cases, spikes in nighttime lighting at a specific location represent new electricity consumption where little to no nighttime lighting previously existed. Figure Three below provides overhead satellite imagery and nighttime lighting data for such a facility in Turpan Prefecture, in northeastern Xinjiang. It shows a high security reeducation center under construction in 2018 in a formerly empty rural area, with building underway in the southeast corner of the property, a large volume of cars visible in a parking lot in the northeast corner, and construction storage visible in the northwest corner.

Construction as shown by the nighttime lighting data in this figure grew rapidly in early 2017, and approached completion in early 2018. The growth period flagged by our double moving average (shaded in gray) leads these dates by a few months, while the decline period flagged by our algorithm (shaded in yellow) captures the point at which construction of this facility began to slow in pace.

Figure Three. Detention Facility Construction in Rural Area of Turpan Prefecture, Xinjiang

In other cases, spikes in nighttime lighting offer evidence of major construction at preexisting schools, prisons, or other facilities that are repurposed to serve a different role, or simply expanded to handle a larger volume of detainees. Figure Four below provides overhead satellite imagery and nighttime lighting data for such a facility in Ili Prefecture, in northwestern Xinjiang. A low-security reeducation center appears to have been constructed in place of an existing school in a built-up urban area. Major construction of a new building within the center of the facility is seen in satellite imagery from September 2018, at roughly the same time as a major spike in nighttime lighting.

In this case, our nighttime lighting analysis flags the start of growth at this facility in early 2017, based upon an inflection point in the smoothed moving average. This roughly correlates with the incidence of major construction at this facility seen in satellite imagery, although we lack access to imagery from the exact start of this growth period. Similarly, overhead imagery shows few signs of construction following the start of the decline period flagged by our algorithm, beginning in early 2019.

Figure Four. Detention Facility Construction at Existing Facility in Ili Prefecture, Xinjiang

Overall then, we assess that major spikes in nighttime lighting represent evidence of construction at known detention facilities. In some cases, this growth represents construction of new facilities from scratch, or even new buildings on facilities as they expand over time. In other cases, it represents construction activity to repurpose existing facilities to serve as detention camps. Either way, nighttime lighting serves as a useful metric to capture both the timeliness and magnitude of growth in Xinjiang's broader detention system.

Measuring the Overall Growth of Detention Facilities

After analyzing the nighttime lighting trends of all 380 detention facilities in our dataset, we can ultimately assess the extent to which major periods of construction occurred across the broader Xinjiang detention system. Figure Five below plots the number of detention facilities in our dataset experiencing a major period of growth in nighttime lighting for each month since January 2014. We include those facilities experiencing a major decline as well, for reference. In a given month on this figure, each bar represents the number of distinct detention facilities experiencing an uninterrupted period of growth (blue) or decline (red) in smoothed nighttime lighting.

Figure Five. Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, 2014-2020

This provides a clear, visual representation of the massive growth in Xinjiang's Uyghur and ethnic minority detention system over time. Importantly, it reveals a clear inflection point beginning in 2016, when the construction of detention facilities across Xinjiang ballooned at a rapid pace. Throughout 2016, an average of 15 new facilities per month entered into new, sustained stretches of growth in nighttime lighting. By 2017, construction across Xinjiang would reach its overall peak, with an average of 264 detention facilities per month in the midst of sustained growth in nighttime lighting. This rapid pace of construction and expansion would decline only slightly in 2018, with an average of 240 facilities per month in the midst of sustained nighttime lighting growth.

Elevated levels of construction in Xinjiang would continue in earnest through early 2019, at which point the number of detention facilities beginning new periods of growth in nighttime lighting falls considerably. This does not suggest, however, that construction of detention facilities came to a halt in early 2019. Rather, this analysis highlights that construction continued within nearly 80 facilities per month through at least mid-2020, the endpoint of our analysis.

This finding comports with existing evidence based upon satellite imagery, as compiled in ASPI's original investigative work, to suggest that at least 61 detention facilities have been constructed or expanded between July 2019 and July 2020. It also remains possible that new facilities have been constructed elsewhere in Xinjiang not captured in ASPI's dataset of 380 facilities used in this analysis.

Either way, our analysis suggests that construction of Xinjiang's broader detention system was primarily concentrated over a roughly 36-month campaign, beginning in 2016 and lasting through early 2019. Over this time period, 69 percent of detention facilities in our dataset began major growth periods in nighttime lighting, lasting an average of 27 months, and ending their growth periods with 370 percent higher nighttime lighting levels on average.

More importantly, the stark inflection points marking the beginning and end of this expansion period imply deliberate efforts to ramp up the pace of reeducation, detention, and imprisonment of Xinjiang's Uyghur and ethnic minority population over this time period, in line with existing research.