Unlocking the Secrets of the Angkorian World: AI-Powered Mapping of Ancient Reservoirs

The Challenge of Mapping a Lost Empire

The Angkorian civilization, a powerful force in Southeast Asia from the 9th to 15th centuries, left behind a sprawling legacy. Archaeologists have meticulously mapped the Greater Angkor Region, uncovering thousands of temples, occupation mounds, and a complex network of water management infrastructure. But this is just the tip of the iceberg. The vast landscapes beyond, once part of the Angkorian sphere of influence, remain largely uncharted.

Manually mapping these expansive areas using traditional methods would be a Herculean task, demanding countless hours of painstaking analysis of satellite imagery. This is where the power of artificial intelligence comes into play.

Harnessing Deep Learning for Archaeological Discovery

Our research team employed a cutting-edge deep learning model, DeepLabV3+, to automate the identification of a key Angkorian feature: reservoirs. These vital components of the ancient landscape served multiple purposes, from water storage for agriculture and livestock to providing a source of protein through fishing.

We trained our AI model using high-resolution satellite imagery from Microsoft Bing and a meticulously curated dataset of over 11,000 reservoir shapes from across Cambodia. This process also allowed us to address challenges like cloud cover, modern development obscuring ancient features, and changes in reservoir appearance over time that made some legacy data unusable. This demonstrates the complexities of bringing historical data to life using modern remote-sensing technology.

We augmented the model with a form of semi-supervised learning called 'pseudo-labelling', drastically increasing the automatically curated data used during training for greater sensitivity and accuracy. This allowed the model to learn the unique characteristics of Angkorian reservoirs, even those no longer holding water or with eroded walls. The model's ability to adapt to such nuances was instrumental in maximizing its predictive power.

AI: A Powerful "Second Opinion" for Archaeologists

Our results demonstrate that the AI model, while not a replacement for human expertise, serves as a valuable tool for accelerating the mapping process. In test areas across Cambodia, the model achieved an F1-score of around 31%, correctly identifying a significant portion of known reservoirs.

In a head-to-head comparison with experienced archaeologists and students, the AI model proved to be incredibly fast, completing the analysis of a 130 km² area within the Greater Angkor Region in just two minutes. While the experts achieved higher accuracy, particularly in minimizing false positives, their analysis times measured in minutes and hours dwarfed that of the AI's task-completion time.

A Collaborative Future for Archaeology and AI

The true potential of this technology lies in a collaborative workflow, where AI acts as a powerful "second opinion" for archaeologists. By pre-scanning vast areas and identifying potential reservoir locations, the AI significantly reduces the time and effort required for human verification.

As Verschoof-van der Vaart and Lambers (2021) noted, integrating human and AI workflows increases overall productivity and mitigates the effects of human fatigue. This is crucial for large-scale projects, where maintaining consistent attention to detail can be challenging.

This collaborative approach not only accelerates the mapping process but also enables archaeologists to focus their expertise on interpreting the results and uncovering the intricate stories hidden within the ancient landscape.

Expanding the Horizons of Discovery

Our next steps involve refining the AI model to improve its accuracy and expanding its capabilities to identify other Angkorian features, like temples, and their spatial relationships. We aim to push the boundaries of what's possible, leveraging AI to unlock the secrets of a lost empire and deepen our understanding of ancient urbanism and early statehood in Southeast Asia.

The integration of AI into archaeological research holds immense promise. By combining the strengths of human expertise and machine learning, we can accelerate the pace of discovery and gain a richer, more comprehensive understanding of our shared past.