A3RD (AI-Assisted Archaeological Remains Detection)

As part of its ongoing efforts to push the boundaries of archaeological research, CAMEL has launched a collaborative project that revolutionizes how archaeologists and cultural heritage professionals detect and map features across large landscapes: the AI-Assisted Archaeological Remains Detection (A3RD) project.

Our project aims to develop the first open-source, AI-based remote sensing platform designed specifically for archaeology and the humanities. The project focuses on building a scalable, user-friendly deep learning ecosystem that makes AI-assisted feature detection transparent, reproducible, and accessible—particularly for researchers with little to no technical background in programming or data science.

The Problem

The Earth’s surface is covered with traces of human history—settlements, infrastructure, and agriculture. With the advancement of remote sensing technology—the use of satellite and aerial imagery to analyze cultural and environmental features—archaeologists now have powerful tools to study these patterns across vast geographies and longue dureé. Remote sensing has significantly expanded the scale at which archaeological sites can be identified and interpreted. Yet, its full potential remains constrained by manual, labor-intensive workflows that are time-consuming, difficult to scale, and expertise reliant.

At the same time, while deep learning has transformed industries from healthcare to finance in the past decade, archaeology and cultural heritage management have yet to fully benefit from these technological advances. Manual interpretation of imagery remains the norm, largely due to significant barriers to adopting AI.

One key challenge is that geospatial AI poses domain-specific complexities beyond those of conventional computer vision. These include non-standard file formats, embedded geospatial metadata, varying projection systems, massive file sizes, and the use of multispectral or hyperspectral imagery. Beyond the technical obstacles, there’s a skills and training gap, most archaeologists lack formal education in programming, data science, or DevOps, making it difficult to use or adapt existing open-source AI tools within scalable research workflows. Moveover, unlike sectors such as agriculture or weather forecasting, where commercial and governmental incentives drive investment in geospatial AI, archaeology lacks commercial values. As a result, the tech industry has devoted far less attention to developing AI solutions for cultural heritage, despite its immense cultural, educational, and scientific value.

By democratizing access to AI-powered remote sensing, A3RD empowers archaeologists, cultural resource managers, and scholars to work at greater scale and efficiency—unlocking new forms of research and global collaboration in the process.

Our Solution

A3RD bridges the gap between archaeology and cutting-edge AI by building an open, flexible, and user-friendly ecosystem for geospatial deep learning. The platform integrates modern geospatial standards, such as SpatioTemporal Asset Catalog (STAC), with widely used open-source tools to support every stage of the AI workflow. These include tools for data labeling and validation using QGIS, model versioning and registry through MLflow and Hugging Face, as well as training and performance monitoring with MLflow and TensorBoard. All components are supported by comprehensive technical documentation to guide users throughout the process.

The full workflow is designed to function seamlessly across a variety of computing environments, including personal desktops, the University of Chicago’s high-performance computing cluster Midway3, and commercial cloud platforms such as AWS. This flexibility ensures accessibility for researchers with varying levels of infrastructure and resources.

To support education and global collaboration, CAMEL will maintain open access to training imagery, models, and published outputs. Most importantly, to make AI-assisted detection truly accessible, A3RD integrates its models and workflows directly into QGIS as plugins. This allows archaeologists and researchers already familiar with this widely used open-source platform to engage with advanced AI tools—no coding required.

Our Team

The A3RD team is based at the University of Chicago and brings together an interdisciplinary group of experts—spanning archaeology, geospatial science, computer vision, and digital humanities—from multiple institutions and research centers around the world.

• Prof. Mehrnoush Soroush: A3RD Principal Investigator and Domain Lead in Archaeology and Archaeological Remote Sensing.
  Assistant Professor of Ancient Near Eastern Studies, Director of CAMEL Lab, University of Chicago
• Prof. Andrew Wilson: Domain Expert in Archaeology and Cultural Heritage Management.
  Professor of the Archaeology of the Roman Empire, Senior Researcher at Endangered Archaeology in the Middle East and North Africa (EAMENA), University of Oxford.
• Dr. Bijan Rouhani: Domain Expert in Archaeology and Cultural Heritage Management.
  Senior Researchers at Endangered Archaeology in the Middle East and North Africa (EAMENA), University of Oxford
• Rémi Cresson: Project Technical Lead, Lead Developer of OTBTF Software.
  Senior Geospatial Engineer at French National Institute for Food, Agriculture, and Environment (INREA)
• Dr. Nick Ross: Senior Data Scientist and Software Architect.
  Data Science Clinic Director at Data Science Institute, University of Chicago
• Dr. Parmanand Sinha: High Performance Computing Scientist and Geocomputational Researcher.
  Computational Scientist at Research Computing Center (RCC), University of Chicago
• Dr. Emad Khazraee: Product Strategy Lead.
  Vice President of Data Science & AI at Xometry Inc.

Research assistants in this project include:

• Dominik Lukas: Student Research Team Lead.
  PhD Candidate in Anthropology, University of Chicago
• Yuwei Zhou: Student Research Team Lead.
  PhD Candidate in East Asian Languages and Civilizations, University of Chicago
• Çağlayan Bal: Senior Student Research Assistant.
  PhD Student in Middle Eastern Studies, University of Chicago
• Jiayue Wang: Programming and Data Science Student Researcher.
  Undergraduate Student, University of Chicago

Read more about who we are and what we do on the team page.

Activities & Outcomes

With seed funding from ISAC, CAMEL hosted a series of workshops and hackathons from winter through spring 2024, designed to give student researchers hands-on experience with machine-learning and deep-learning methods. Through a mix of lectures and practical tutorials, students learned how to build and deploy AI models that could assist in archaeological feature detection using satellite imagery.

From March 16th to 23rd, 2025, the A3RD project held a week-long workshop at the University’s John W. Boyer Center in Paris. Funded by a Faculty Grant from the University of Chicago’s International Institute of Research in Paris (IIRP), the workshop brought together key international collaborators to advance the next phase of A3RD’s global qanat detection initiative.

Case Studies

Detecting Qanats from Space: An A3RD Case Study

What are Qanats and Why Do We Care?

A3RD’s initial test case focuses on the detection of qanats, a sophisticated, ancient subterranean water-management system widespread historically across the arid and semi-arid regions of North Africa, the Middle East, and Central Asia. These gently sloping tunnels, accessed through vertical shafts, were engineered to tap groundwater and transport it over long distances without the need for pumps. For thousands of years, qanats enabled agriculture, trade, and long-term settlement in some of the world’s driest environments, making them one of the most remarkable technological innovations in human history.

Many qanats are still in use today, continuing to supply water for irrigation and daily life. Their maintenance has traditionally relied on local knowledge and specialized skills passed down through generations. Mapping qanats—documenting their location, length, direction, and condition—not only helps us understand ancient technologies and human adaptation to harsh climates, but also supports the preservation of cultural heritage and practical water-management knowledge that is still relevant today.

However, qanats are disappearing at an alarming rate. A recent study shows that less than 6% of the qanat shafts in 1961 remained intact in 2021—a dramatic decline caused by rapid development, neglect, and climate change. This loss has created an urgent need for archaeologists, heritage professionals, and local communities to document surviving qanats before they vanish entirely.

From CORONA Imagery to Scalable Insights

Currently, the most effective way to locate qanats is through manual inspection of CORONA satellite imagery—a Cold War-era archive (1960–1972) widely used in archaeology. This combination is uniquely valuable for our target users–unlike other fields of research using contemporary satellite data, archaeological remote sensing research primarily depends on historical imagery (1950s–1980s), as many cultural landscapes have since been lost to modern development. On these historic images, qanats appear as lines of evenly spaced soil mounds (spoil rings), resembling a string of Lifesaver candies.

Although CORONA data has helped identify qanats across regions from Morocco to western China, our understanding of qanats remains fragmented, based largely on manual inspection on RS data and isolated case studies. A comprehensive understanding of qanat-building cultures and geographies and the technology’s impact on socio-economic resilience requires tools that can operate at scale, across diverse landscapes.

Promises and Challenges in AI-Assisted Detection of Qanats

A3RD is using declassified CORONA satellite imagery to automate the detection of qanats across the regions where they were historically built. Our goal is to create the first global digital dataset of qanats: a high-resolution, publicly accessible map that will support research in archaeology, history, and environmental studies in exploring how qanat technology shaped long-term patterns of settlement, agriculture, and water use in some of the world’s most arid regions. It will also assist local communities and heritage professionals in preserving qanats—many of which are still vital for water access today.

From a computer vision perspective, automated qanats detection isn’t straightforward as it seems. While their vertical shafts leave circular spoil mounds that appear in relative distinctive and regular lines on satellite images, the appearance of qanats varies greatly depending on local geography, image quality, and how the system was built and maintained. This diversity in form and context poses a challenge for standard AI tools, which often perform best when looking for consistent, clearly defined patterns.

To overcome this, A3RD is developing AI models that learn from a variety of context-specific examples across different landscapes and cultural settings, in order to diversify training data as much as possible. At the same time, we’re exploring multi-modal AI approaches that go beyond just image recognition, incorporating descriptive information during the labeling and evaluation process. This will make the models more flexible, robust, and applicable to the wide range of qanat systems found across Eurasia.