The process has been developed from the template of the intelligence cycle to suit the needs of this investigation. Multiple phases can run simultaneously for different prisons, however the process of confirming a single prison follows these phases sequentially.
Phase 1 – Data Collection
- Manual location of prisons and possible detention facilities in towns, cities, villages, and remote areas of Tibet. This will be done by broad searches of the area’s satellite imagery, as well as being guided by any intelligence from section b.
- Obtaining all documents and references to prisons and detention facilities and bringing them to attention.
- Image collection for verification and leads. Including SOCMINT.
- Witness collection for verification and leads.
Phase 2 – Data Refinement
- Connecting known prisons to their coordinates and noting key features such as perimeter, area, and number of watchtowers.
- Correctly identifying (definite) prisons and if name unknown, ensuring they have a unique ID.
- Identifying possible prisons and detention centres and marking them as “possible” for additional research.
- Identifying references to facilities without known candidates and referring back to Phase 1.
Phase 3 – Analysis
- Identifying all key features which will help inform future iterations of Phase 1 and Phase 2.
- Obtaining key information on all prisons and detention centres, such as perimeter, area, and number of watchtowers.
- Verifying each facility identified
- Identifying false positives and archiving them.
- Establishing key dates, such as date built, date(s) expanded, and date demolished.
Phase 4 – Redacting Personal Details and Ethical Evaluation
Ensuring the document is safe to distribute in Phase 5. This involves removal of any sensitive details [limited to personal identifiable information] which may compromise individuals or sources. This phase may not be needed and will not be documented. Note that closed-source information has not been used so this should not impact integrity of data.
Phase 5 – Crowdsourcing Scrutiny
- Opening up the research to online contributors in a controlled manner to allow for scrutiny and peer analysis.
- Addressing all feedback and revisiting details which are flagged as concern points.
Phase 6 – Review
- Compiling the research and running additional checks on all data found.
- Sending the complete document and dataset to be reviewed by individuals and organisations who may be interested.
Performed Activities
This section aims to provide an overview of the way in which locations were discovered and analysed. There were multiple ways in which the data was obtained, each of which is outlined below.
Locating and identifying facilities
Process Flow 1
| Imagery search ➔ Classification ➔ Identification ➔ Verification ➔ Submission |
Process Flow 1 was the most common process for obtaining information in this project. Many of the locations were identified by satellite imagery searches which involved scanning Google Earth for possible locations and marking them down. These findings were then classified as detention facilities, possible hits, or rejected. From here attempts to identify the location involved searches online for references to detention in that region. If matches were found, they were verified and submitted to the report.
Process Flow 2
| Imagery search ➔ Classification ➔ Failed identification ➔ Submission as “Unidentified” |
When locations could not be identified, they were submitted as “unidentified” locations. The analysis was cyclical and these locations could be picked up later in the research as new information arose.
Process Flow 3
| Information obtained ➔ Check against “Unidentified” locations ➔ Identification ➔ Verification ➔ Submission |
Process Flow 2 covers the direction of analysis when information was obtained about a detention facility first. Whenever any new information was found online regarding locations, it was referenced against the list of unidentified locations to see if there were any matches. Following this, the findings were verified against known information and imagery and then submitted to the report.
Process Flow 4
| Information obtained ➔ Information collection ➔ Search imagery ➔ Identification ➔ Verification ➔ Submission |
Process Flow 3 overviews the flow when information was obtained about a location but there was no match against locations discovered in the project. This method relied on additional research to determine where to start the search, the search criteria, and any clues about what to look for. Once a location was determined, it then went through the same verification process to ensure findings were accurate, before it was submitted to the report.
Online data gathering
Obtaining information from open sources online was key to verification and location identification. The process involved identifying every name for a location and running searches in various search engines to find all mentions and images. From there, information was aggregated and sources evaluated. For many sources, the quality of information was poor and so they were flagged for additional investigation and verification.
One of the most prominent search engines used was Baidu, which allowed better searches of Chinese media and government websites. The process for Baidu included translating prison names into Chinese and finding alternative names. When suggested searches were relevant, they were also followed to produce additional leads or search terms to broaden the search radius and counter some of the selection biases inherent in search engine use.
All images found were saved and allocated to folders. Images deemed to be useful were flagged for geolocation and verification.
Image analysis and verification
Image analysis was a crucial task in the identification and verification of facilities. It involved standard methods of highlighting key features and obtaining additional imagery to perform comparisons and determine matches. Due to the variation in dates that images may have been captured, analysis was performed on both contemporary and historical satellite imagery.
Tools such as Photoshop and other image editing software were used to highlight the analysis and make connections between ground-level imagery and satellite imagery. Some images were enhanced or lightly edited to improve visual clarity.
Another key component of satellite imagery analysis was to determine the dates of construction, expansion, and decommissioning of facilities. For this, historic satellite imagery was used and dates of change were noted. Due to the poor date resolution of historic satellite imagery, some data entries are given as ranges. This information could also be referenced against known dates of construction or expansion for additional verification of the locations.