In the first report in this series, RAND introduced its approach to empirically assess the rapid, coordinated growth of Uyghur and ethnic minority detention facilities across Xinjiang using nighttime lighting. Our analysis revealed a clear inflection point beginning in 2016, after which the pace of growth in Xinjiang's detention system accelerated rapidly, and continued unfettered for roughly thirty-six months to follow.

In our second report, RAND employed a new approach based upon these nighttime lighting data to systematically assess the operating status of known detention facilities. This analysis revealed that only three percent of known detention facilities in Xinjiang show clear evidence consistent with likely closure, despite official Chinese government claims to the contrary. Instead, nearly half of all Uyghur and ethnic minority detention facilities analyzed across Xinjiang exhibited uninterrupted growth in nighttime lighting since their construction.

This final report dives even deeper into the Xinjiang detention system, exploring the broader trends in growth and decline of facilities across Xinjiang based upon their known characteristics such as its type, location, proximity to residential areas and other facilities, and visible security features. It reveals new insights into the types of facilities, and geographic locations, where growth in nighttime lighting has continued through 2020. This research aims to help future researchers chart the trajectory of China's widespread detention of Uyghur and ethnic minority populations in the region going forward, in part, by understanding where it is headed today.

Background and Methodology

To assess broader trends in the growth of China's detention system in Xinjiang, we return to our nighttime-lighting-based assessment of the current operating status of detention facilities in Xinjiang, as presented in the first two reports in this series. Overall, our approach calculates monthly nighttime lighting data over known detention facilities in Xinjiang, based on a dataset of 380 known or suspected facilities originally identified and geolocated by the Australian Strategic Policy Institute (ASPI). Using the locations of these facilities, we measure the average nighttime lighting over each facility on a monthly basis from 2014 through mid-2020 and identify major periods of uninterrupted growth or decline in nighttime lighting over time. Further details on this methodology are provided in our first report.

Based upon these nighttime lighting data, we then identify a set of four common growth trajectories that appear consistently across nearly all of the 380 detention facilities in our dataset. These growth trajectories include those detention facilities with sustained growth in nighttime lighting over our entire time period, those with stable or cyclical nighttime lighting over time, those with major growth followed by a partial decline, and finally, those with major growth followed by a near-complete decline in nighttime lighting. Further details on this methodology are provided in our second report.

Overall, we classify each of the known detention facilities in Xinjiang into one of these four growth trajectories, as summarized below in Table One. This table also provides a visual example of the nighttime lighting trajectory commonly seen in each category.

Table One. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, 2014-2020

Our previous report focused primarily on those facilities in the fourth growth profile, which experienced major growth in nighttime lighting followed by a significant decline to levels at or near pre-construction nighttime lighting. Using overhead satellite imagery, we then individually assessed the operating status of each facility, determining that only 11 of these facilities showed evidence of likely closure. While many of the remaining 40 facilities in this category showed clear evidence in satellite imagery that they remained open, others proved inconclusive.

Beyond the nighttime lighting data on the 380 facilities in our dataset, we also possess data on various characteristics of each facility. This includes the type of facility (e.g. reeducation center versus detention center or prison), as well as its location, proximity to residential areas and other detention facilities, as well as the presence of visible security features such as interior fencing, external walls, or watchtowers. The vast majority of these data were gathered by ASPI in its original investigation of these facilities.

Using these data, this final report seeks to explain variation in nighttime lighting across these known characteristics, in order to better understand which detention facilities are likely to expand or decline over time going forward. It focuses on the relationship between these characteristics and the four broad growth profiles identified using nighttime lighting data in our prior report, as general indicators of the growth or decline in activity at detention facilities writ large.

For each covariate describing potential differences in nighttime lighting across detention facilities in our dataset, this report explores two broad questions:

1. First, are certain types of detention facilities more likely to experience sustained growth, stable, partial decline, or major declines in nighttime lighting?
2. Second, have certain types of detention facilities experienced major growth or decline at different points over time, particularly in recent years? 

Type of Facility

We begin by assessing whether the nighttime lighting signature of detention facilities in Xinjiang differs based upon the type of each facility. Recall that ASPI categorizes each detention facility in our dataset into four distinct tiers, based upon their imagery signature and visible level of security. Of note, we adjust these tiers to account for our alternative assessment of the imagery signature of administrative detention centers relative to long-term prisons (described in further detail in our first report). As such, we classify the various facilities in our dataset as either low-security reeducation centers (Tier One), high-security reeducation centers (Tier Two), long-term prisons (Tier Three), and administrative detention centers (Tier Four).

Figure One below plots the distribution of our four nighttime lighting growth trajectories based upon the type of each detention facility in our dataset. Those facilities with uninterrupted growth in nighttime lighting are shown in blue. Facilities with stable or cyclical nighttime lighting are shown in green. Those with a partial decline in nighttime lighting are shown in yellow, and facilities with major declines (indicating possible closure) are shown in red.

Figure One. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, 2014-2020, by Type of Facility

Overall, this analysis reveals interesting differences between the growth patterns of lower-security reeducation centers as compared to higher-security prisons and administrative detention centers. On average, reeducation facilities more frequently showed partial or major declines in nighttime lighting over time. Conversely, long-term prisons and administrative detention centers have shown much more frequent growth in nighttime lighting writ large, at least over the course of the time period we analyze this research (2014 to mid-2020).

Figure Two below takes a different approach to understand these trends. Rather than focus on the general growth trajectory of each type of facility over the entire dataset, it focuses instead on when these facilities experienced major growth or decline over time. Figure Two plots the percentage of all facilities by type experiencing a major period of growth (positive values) or decline (negative values) in nighttime lighting, for each month since January 2014.

Figure Two. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Type of Facility

Several trends emerge in this figure. First, it appears that low-security reeducation centers (Tier One) experienced slightly lower levels of major growth in nighttime lighting between 2016 and 2018 compared to other types of facilities. This is somewhat surprising, given that our prior research (in the first report of this series) suggested that the majority of construction across Xinjiang occurred over this time period. However, reduced growth in nighttime lighting could suggest that low-security reeducation centers facilities were more likely to be repurposed from existing infrastructure, or that they took less time to construct relative to more secure facilities.

These results also suggest that long-term prisons (Tier Three) have become a primary point of emphasis in recent years, with a comparably high percentage of such facilities showing active growth in nighttime lighting in 2019 and 2020 compared to other types of facilities. This suggests that imprisonment of Uyghur and ethnic minority groups in Xinjiang has taken a greater priority than reeducation in recent years.

Region

To get a better sense of the geographic distribution of detention facilities experiencing growth and decline in nighttime lighting, we next compare the growth profiles of detention facilities in northern Xinjiang relative to those in southern Xinjiang. Historically, northern Xinjiang has been home to a larger percentage of Han Chinese than the south, many of whom have deliberately resettled into Xinjiang with support from the Chinese government. Moreover, northern prefectures in Xinjiang have a higher population of Mongol, Kazakh, and Kyrgyz minorities in addition to ethnic Uyghurs. Conversely, southern Xinjiang has been the historical heartland of Xinjiang's Uyghur population. Differences between the two regions could suggest a greater emphasis on detaining certain minority groups over others, or shifts in local preferences to manage large-scale detention facilities.

Figure Three below plots the distribution of our four common nighttime lighting growth trajectories between northern and southern Xinjiang, divided based upon prefecture as described in the note to this figure. Overall, fewer facilities in southern Xinjiang experienced uninterrupted growth in nighttime lighting over time compared to the north, and more southern Xinjiang detention facilities experienced partial declines in nighttime lighting. But overall, these differences appear relatively marginal at first glance.

Figure Three. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Region

To get a sense of the timing of growth in detention facilities in each region, Figure Four below plots the frequency of detention facilities experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time across northern and southern Xinjiang. This analysis reveals greater differences between the two regions. Overall, it appears that detention facilities in southern Xinjiang were not just more likely to experience uninterrupted growth in nighttime lighting over time, but they also started to expand much earlier, and at a faster rate than facilities in northern Xinjiang. Since 2018, however, the growth profiles of detention facilities in both regions appear to behave similarly.

Figure Four. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Region

Major Prefectures

Acknowledging these regional differences, we next focus on whether any specific prefectures in Xinjiang experienced more frequent growth or decline in nighttime lighting than others. We focus this analysis on just six prefectures, each with more than 25 separate detention facilities, but that collectively account for roughly two-thirds of all detention facilities in Xinjiang.

Figure Five below shows the location of these facilities across these prefectures. The heaviest concentration of detention facilities in Xinjiang, by far, is in the Uyghur heartland of Kashgar and Hotan, with 72 and 52 facilities each, respectively. To the north along the Kyrgyzstan border, Aksu Prefecture holds an additional 32 detention facilities. Even further north, along the Kazakhstan border, Ili Prefecture has an additional 46 facilities. To the southeast, Bayingolun Prefecture occupies much of the center of Xinjiang, as well as its land borders with neighboring Qinghai province and the Tibet Autonomous region. It houses an additional 27 known detention facilities. Finally, the prefecture-level capital city of Xinjiang, Urumqi, has an additional 25 facilities within its borders as well.

Figure Five. Prefectures in Xinjiang with 25 or More Detention Facilities

While these six prefectures represent the vast majority of Xinjiang's broader detention system, their respective detention facilities show largely different nighttime lighting signatures over time. Below, in Figure Six, we show the distribution of nighttime lighting growth trajectories within each prefecture.

Figure Six. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Prefecture

Detention facilities in Kashgar, Hotan, and Aksu prefectures exhibit relatively comparable nighttime lighting growth trajectories, with detention facilities in Aksu showing slightly elevated levels of uninterrupted nighttime lighting growth. Northwestern Ili Prefecture stands out from the pack, with a significantly greater incidence of detention facilities with major declines in nighttime lighting. Conversely, detention facilities in the capital city of Urumqi showed incredibly stable nighttime lighting signatures over time, with a drastically lower incidence of detention facilities with declining levels of activity compared to other prefectures.

Figure Seven explores the timing of growth and decline in nighttime lighting across detention facilities in these prefectures. As shown below, this figure plots the frequency of detention facilities in each of these prefectures experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time. Most prefectures appear to have followed a similar growth trajectory over time, with two major exceptions. Specifically, nighttime lighting growth at detention facilities in Ili Prefecture and in Urumqi occurred much more slowly than elsewhere through much of Xinjiang's expansionary period from 2016-2017.

Figure Seven. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Prefecture

But by early 2018, nighttime lighting in Ili would rise rapidly to levels comparable to other prefectures, and fall drastically through the beginning of 2019. In Urumqi, however, nighttime lighting growth would remain relatively infrequent throughout much of 2018, but then continue into 2020 at much higher rates than seen elsewhere in the province.

Xinjiang Production and Construction Corps (XPCC)

Beyond these major prefectures, we can also isolate the nighttime lighting signatures of those detention facilities located in cities administered by the Xinjiang Production and Construction Corps (XPCC). The XPCC, known simply as the Bingtuan (or corps) in Mandarin, is a state-owned industrial and paramilitary organization first established in 1954 in order to stabilize and develop Xinjiang, partially in light of its restive history. Over the years, the XPCC has built and developed a number of sub-prefectural cities across Xinjiang, which operate autonomously and independently from the broader Xinjiang civil administration.

ASPI's underlying data on known Xinjiang detention facilities suggest that the XPCC operates at least 21 separate facilities across 7 sub-prefectural cities, including Aral, Beitun, Shihezi, Shuanghe, Tiemenguan, Tumshuq, and Wujiaqu. Figure Eight below plots the distribution of nighttime lighting growth trajectories over known detention facilities in XPCC-administered cities, compared to those elsewhere in Xinjiang. The differences between these two groups appear stark, suggesting that detention facilities operated by the XPCC have seen much higher rates of growth and expansion, and minimal periods of decline in nighttime lighting over the past several years.

Figure Eight. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Location in XPCC-Administered City

Figure Nine below plots the frequency of detention facilities in XPCC-administered cities experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time. These results are equally stark, suggesting that XPCC-administered detention facilities continued to expand with much greater frequency into 2019, well past the point at which most detention facilities elsewhere in Xinjiang began to show major declines in nighttime lighting.

Figure Nine. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Location in XPCC-Administered City

Proximity to Residential Areas

We can also assess the extent to which nighttime lighting growth has varied between detention facilities in urban and rural areas of Xinjiang. To do this, we use ASPI's original data on the proximity of each detention facility to nearby residential areas. While not a perfect proxy for urban-rural differences, this metric does help to assess whether greater emphasis has been given to shifting Xinjiang's detention apparatus away from major population centers and into more remote locations.

Figure Ten below plots the growth profiles of detention facilities based upon their proximity to residential areas, as gathered by ASPI. Of note, ASPI's underlying data includes a large volume of missing data for this measure. However, these trends still prove illustrative, suggesting that detention facilities in closest proximity to residential areas (within one kilometer) have experienced significantly greater declines in nighttime lighting compared to other facilities.

Figure Ten. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Proximity to Residential Areas

To assess the timing of these effects, Figure Eleven below plots the frequency of detention facilities experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time, by proximity to nearby residential areas. Few new insights emerge from this figure, except to affirm that those facilities further away from residential areas experienced more frequent growth than others, particularly from 2016 to 2018.

Figure Eleven. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Proximity to Residential Areas

Proximity to Other Detention Facilities

Across Xinjiang's broader detention system, it is common to see detention facilities collocated with other detention facilities that often serve a different purpose (i.e. of a different type). For context, the average detention facility in our dataset is located only 6.4 kilometers away from the next closest facility on average. But roughly half of all detention facilities are located less than one kilometer from the next closest facility, often within the same compound. 

Figure Twelve below shows overhead imagery of an example multi-facility detention complex in Ili Prefecture, with a low-security reeducation center (shaded blue) sharing a larger compound with an administrative detention center (shaded green). Other buildings on the compound serve unknown purposes, but overall, these two separate facilities share a common exterior wall, front gate, and road access to the north. Any potential economies of scale from collocating such facilities could incentivize greater physical consolidation of facilities across Xinjiang's detention system going forward.

Figure Twelve. Example Multi-Facility Detention Complex in Ili Prefecture, Xinjiang

To assess whether physical proximity correlates with the overall growth trajectory of specific facilities, Figure Thirteen below plots the distribution of growth profiles by distance to the next closest detention facility. Overall, these results suggest only minor differences in growth trajectory, with closely proximal (or collocated) facilities demonstrating marginally higher frequencies of uninterrupted growth relative to facilities located further apart. Surprisingly, those facilities located more than 10 kilometers from the next closest facility show similarly high levels of uninterrupted growth.

Figure Thirteen. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Proximity to Other Detention Facilities

To better gauge, the timing of these effects, Figure Fourteen below plots the frequency of detention facilities experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time, based upon their proximity to other detention facilities. Again, we see that facilities both in close proximity to others, and those located more than 10 kilometers away from the next closest facility, experience similar growth patterns over time. Nevertheless, collocated facilities appear to have experienced a decline in nighttime lighting more frequently in recent years.

Figure Fourteen. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Proximity to Other Detention Facilities

Evidence of Desecuritization

In its original analysis, ASPI also assessed whether detention facilities showed signs of desecuritization including through the removal of watchtowers, interior fencing, or exterior walls meant to provide heightened security for certain detention facilities. Reduced levels of security over time could suggest that a detention facility has been downgraded to become a different type of facility within Xinjiang's detention system, such as transitioning from a high-security reeducation center to become a low-security reeducation center. Alternatively, downgraded security could imply that such facilities have actually closed, or been converted into buildings that serve an alternative civilian purpose.

Across the 380 facilities in its dataset, ASPI occasionally lacked sufficient imagery to assess evidence of desecuritization. As such, roughly one-third of the facilities in our analysis have missing data for this variable. However, the nighttime lighting growth trajectories of the remaining facilities still prove illustrative for our purposes. In Figure Fifteen below, we compare these growth profiles across those facilities with visible evidence of desecuritization, and those without such evidence. The results show a striking difference between these two types of facilities, where desecuritized facilities experienced far higher rates of partial and major declines in nighttime lighting over time.

Figure Fifteen. Nighttime Lighting Growth Trajectories of Xinjiang Detention Facilities, by Evidence of Desecuritization

Figure Sixteen below then plots the frequency of detention facilities experiencing major periods of growth (positive values) or decline (negative values) in nighttime lighting over time, based upon whether or not a detention facility showed visible signs of desecuritization in ASPI's data. The results suggest that nighttime lighting over desecuritized detention facilities began to rapidly decline relative to other detention facilities, particularly in 2019 and 2020. Overall, these findings suggest that desecuritization may serve as a leading indicator of reduced activity and possible closure of known detention facilities.

Figure Sixteen. Xinjiang Detention Facilities Experiencing Major Growth or Decline in Nighttime Lighting, by Evidence of Desecuritization

Other Factors

We also assessed whether nighttime lighting of detention facilities varied based upon the presence of factories inside or adjacent to detention facilities in Xinjiang. As ASPI laid bare in March 2020, a significant aspect of China's detention apparatus in Xinjiang is to use detainees as forced labor at factories, often located on the premises of detention facilities. In its original dataset, ASPI noted whether satellite imagery showed signs of a factory operating inside or adjacent to each detention facility. Surprisingly, we found only insignificant differences in the nighttime lighting signature of detention facilities based upon the presence of these factories. This is all the more surprising given that industrial facilities may themselves produce greater nighttime lighting signatures than residential infrastructure. But is possible such factories do not operate at night, rendering nighttime lighting less useful in this case.

Similarly, we find only marginal differences in nighttime lighting across detention facilities based upon the presence of internal fencing, exterior walls, or watchtowers, as cataloged by ASPI in its original data. More generally, the presence of these security features is heavily correlated with the overall facility type as analyzed earlier in this report, a more useful metric.

Key Takeaways

Overall then, this report sought to assess whether certain factors could help explain the growth and decline of detention facilities across Xinjiang. Key takeaways from this analysis include the following, broken down by each relevant factor:

• Type of Facility: Across Xinjiang, high-security facilities focused on imprisonment and detention of Uyghur and ethnic minority groups have expanded more frequently than lower security reeducation centers. This trend has only accelerated in recent years since 2019, particularly amongst long-term prison facilities rather than administrative detention centers.
• Region: Detention facilities in southern Xinjiang appear to have grown faster and more frequently in the early years (2016-2017), compared to facilities in northern Xinjiang.
• Prefecture: Most detention facilities in Xinjiang are concentrated in a set of six key prefectures across Xinjiang (Kashgar, Hotan, Aksu, Ili, Bayingolun, and Urumqi). Of these prefectures, detention facilities in northwestern Ili prefecture (along the Kazakh border) have experienced far greater rates of decline in nighttime lighting over time. Additionally, Urumqi, the capital of Xinjiang, has shown surprisingly stable levels of activity over time with little evidence of decline in nighttime lighting at its detention facilities.
• Role of the XPCC: Detention facilities in cities administered by the Xinjiang Production and Construction Corps experienced much more robust growth in nighttime lighting over time than facilities elsewhere in Xinjiang, particularly since 2019.
• Proximity to Other Infrastructure: Nighttime lighting data suggest higher rates of growth at facilities further from residential areas, and at those facilities in close proximity to others. However, more recent emphasis has been placed on expanding detention facilities in more isolated locations.
• Camp Characteristics: Visible evidence of reductions in security at a detention facility over time is closely correlated with major declines in nighttime lighting. However, the presence of factories, or initial usage of internal fencing, external walls, and watchtowers at a facility each show little correlation with the nighttime lighting trajectory of detention facilities writ large.