Geodaysit 2023

Andrea Ajmar

Andrea Ajmar, Ph.D., is an expert in the field of Spatial Data Science, geodatabase management and Spatial Data Infrastructure (SDI) development and implementation. He has a solid expertise in environmental management, gained during working experiences both for public and private sector: this expertise provided the correct user perspective, fundamental in the design of efficient and effective database structures and comprehensive SDI solutions. He was the responsible for the realization of UN WFP SDI, a dedicated infrastructure for emergency management in a complex distributed environment: high-availability and redundancy, federated database structures were therefore firstly introduced inside humanitarian organizations operating worldwide. He currently holds the position of fixed-term researcher at the Politecnico di Torino, with teaching positions in master's degrees and doctoral courses. It is part of the laboratory called SDG11Lab which aims to create a spatial data infrastructure capable of guaranteeing efficient access to data sources of general interest (e.g. satellite and airborne acquisitions, reference cartographic data, etc.) and to integrate them with specific thematic data, providing analysis, representation and visualization tools capable of responding to the needs of research teams.


Automatic analysis of detention camps in Xinjiang (PRC) using Nighttime Light remote sensing data
Andrea Ajmar, Edoardo Vassallo, Emere Arco

Global nighttime imaging data, such as Day/Night Band (DNB) VIIRS sensors, provide global daily measurements of visible light and night infrared. Nighttime Light (NTL) remote sensing products have a wide range of applications such as feature detection and monitoring, multitemporal analysis, and prediction of socio-economics and environmental variables.

This work presents a methodology based primarily on NTL data acquired by the VIIRS (Visible Infrared Imaging Radiometer Suite) sensor mounted on Suomi NPP (National Polar-orbiting Partnership) for monitoring the construction of Uyghur’s detention camp in the Xinjiang Uygur Autonomous Region of the People's Republic of China (PRC). This region is strategically important for PRC, with three of the 5 economic corridors of the Belt & Road Initiative (BRI) crossing this administrative unit. Due to its history and culture strongly linked to the Sunni Islamic world and the independence movements rekindled after the dissolution of the Soviet Union, this area is particularly sensitive (and consequently, under special observation) for the Chinese central government. In December 2015, the National People's Congress passed an anti-terrorism law, which defined various aspects of the Uyghur lifestyle and culture as a security issue, contextualizing them as terrorists and extremists.
Since 2014, PRC has begun the construction of detention camps, responding to the first international accusations by denying their existence. Only later, when the existence of the camps was proven more strongly thanks to satellite images and other sources, the Chinese government changed its narrative, by acknowledging their existence only as education camps, intended to help people find stable jobs and improve their lifestyles.
The methodology also exploits day optical images acquired by sensors mounted on Sentinel-2 satellites and data produced by the Xinjiang Data Project that monitors the human rights situation for Uyghurs and other non-Han nationalities in Xinjiang.

Historical series of NTL radiance data has been generated over localities identified as a mass internment camp in a fully automated processing chain based on Google Earth Engine APIs and developed within a Jupiter Notebook, employing also open-source modules.
The procedure works with three major steps:
a) extracts from the database of Google Earth Engine VIIRS nighttime lights data acquired over a list of provided locations and within a user-defined time frame, storing it efficiently;
b) calculates statistics over the radiance values and generates charts displaying the historical trends of the calculated statistical parameters;
c) performs a clustering of the historical series based on Dynamic Time Warping (DTW) and K-Means techniques.
The script has been released and is available on a dedicated GitHub page.

As a result of the procedure, the 380 camps have been grouped into 10 clusters highlighting patterns that can be linked to different phases: construction, operativity, enlargement, dismission, etc. The interpretation of the clusters has been later validated using the visual interpretation of sample Sentinel-2 images and by exploring the relationship between the radiance value and the historical record of the number of buildings within each camp reported in the Xinjiang Data Project dataset.

AIT Contribution
Sala Videoconferenza @ PoliBa