Exploring complexity in the archaeological landscapes of monsoon Asia using lidar and deep learning

Informations projet

archaeoscape.ai

N° de convention de subvention: 866454

DOI

10.3030/866454

Projet clôturé

Date de signature de la CE 7 Mai 2020

Date de début 1 Octobre 2020

Date de fin 30 Septembre 2024

Financé au titre de

• EXCELLENT SCIENCE - European Research Council (ERC)

Coût total

€ 1 518 587,00

Contribution de l’UE

€ 1 518 587,00

Coordonné par

ECOLE FRANCAISE D'EXTREME-ORIENT
France

Periodic Reporting for period 3 - archaeoscape.ai (Exploring complexity in the archaeological landscapes of monsoon Asia using lidar and deep learning)

Période du rapport: 2023-10-01 au 2024-09-30

Résumé du contexte et des objectifs généraux du projet

While it is well recognised that anthropogenic activity has shaped the environment on a planetary scale, our knowledge of historical human impact on Earth's systems remains limited for many parts of the world. This is perhaps especially true of humid tropical environments, where dense vegetation cover typically obscures the traces of human activity that might otherwise be observable on the surface or in conventional aerial and satellite imagery. In places like Southeast Asia, what is left in many cases are the remains of monumental stone structures that hint at the existence of complex societies with large populations, but very little insight into the nature, scale or intensity of their transformations of regional landscapes.

In recent years the use of lidar technology in archaeology has made an important contribution to resolving this problem, revealing faint topographic traces across wide areas, even beneath dense tropical forest. In many areas, lidar has revealed previously-undocumented urban and agricultural landscapes surrounding well-known temple complexes, and in others it has filled in critical lacunae. However, coverage of the landscape often remains patchy, and important areas are yet to be studied. The sheer volume of data makes manual mapping and interpretations of the imagery no longer feasible. Uneven access to large-scale geospatial datasets and the tools required for their interpretation limits effective cross-cultural comparisons and creates barriers to collaboration.

This project has been focused on developing the technical foundations required to address these challenges. It has used cutting-edge lidar technology to undertake the largest-ever acquisition achieved by archaeologists in Asia and distributed the data to stakeholders. Web-based infrastructure that makes these data and the derivative results available to the research community has been developed and continues to be hosted, enabling researchers to overcome systemic barriers and focus on higher-order analytical tasks instead of technical challenges. The project has established a hub for AI-driven archaeo-geospatial analysis, which is pursuing frontier research in computer vision and deep learning, developing computational models for automatically identifying and analysing traces of human activity on the Earth’s surface.

The archives of past activity uncovered with lidar and deep learning offer a unique laboratory for evaluating the nature and degree of human impacts on natural systems from the deep past to the present day, and for tracing pathways of innovation and complexity in the tropical world. Moving forward, the aim is to make a major contribution to a pantropical perspective on how humans have reshaped the surface of our planet in the Anthropocene, with a particular emphasis on trajectories of urbanism, contemporary ecological legacies of past human behaviour, and the ways in which engineered landscapes shape and constrain responses to social and environmental change over the long term.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

Following the unexpected passing of the PI, the funding and duration of the grant were significantly reduced, leading to a contraction in the scope and scale of planned research activities. Nevertheless, the vast majority of the original objectives and ambitions have been successfully achieved.

Over the course of this project, airborne lidar surveys were conducted over several thousand square kilometres of Southeast Asia, bringing remote sensing coverage in the region to a total of more than 7,500 km², embracing key archaeological landscapes in Cambodia, Laos, Thailand, and Indonesia. New data and earlier acquisitions have been integrated into a unified geospatial framework, standardised for analysis, and made available through dedicated web platforms, enabling collaborative work both with established local partners and with the international research community.

A specialised data science laboratory has been established in Paris to support systematic collaboration between archaeologists and AI researchers. This research unit has carried out the effective application of state-of-the-art computer vision models, and developed novel deep learning benchmarks and methodologies. In so doing, it has pushed forward archaeology as an active driver of innovation in machine learning and remote sensing, and demonstrated the generalisability of these models for broader research domains, including such fields as environmental science.

The project has laid the groundwork for an integrated archaeological and ecological agenda, building on its unprecedented-in-scale remote sensing acquisitions in the region along with newly-developed machine learning tools. Initial analysis has started the process of quantifying the development of early societies and the long-term impact and ecological legacy of human activity on tropical environments. Final interpretation, along with the required field verification and targeted archaeological surveys, will be carried out through successor projects.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

Previous archaeological lidar campaigns in Southeast Asia have focused on the areas immediately surrounding well-known monumental remains, introducing spatial bias into datasets and interpretations. A key outcome of this project is the consolidation of a remote sensing dataset covering more than 7,500 km² across diverse archaeological landscapes in Cambodia, Laos, Thailand, and Indonesia, including large swaths of supposed ‘wilderness’ many kilometres removed from known major sites. This has created a baseline for a more objective and systematic understanding of anthropogenic landscape transformation. The data reveal a complex mosaic of settled spaces and early agricultural systems that challenge the conventional divide between ‘natural’ and ‘cultural’ landscapes in the region.

This initiative has marked the first large-scale application of deep learning to archaeological lidar in the Asian tropics, setting a new benchmark through the release of the largest open-access dataset to date. While earlier methods split remote sensing imagery into patches processed independently, this project has re-defined archaeological mapping as a task of semantic segmentation on ultra-high-resolution data. This framing has led to the development of domain-specific hierarchical vision transformer models that effectively combine detailed local information with the broader spatial context. These methodological advances and specialised architectures set a new standard for heritage applications. For the first time, AI techniques developed in an archaeological context have demonstrated generalisability across domains, with promising applications in medical imaging, environmental monitoring, and large-scale geospatial analysis.

The outcomes of this project open new avenues for understanding the deep history of landscape change and environmental history through remote sensing. New evidence gathered through lidar illuminates the structure and evolution of early tropical urban and agricultural systems in mainland Southeast Asia and enables a reassessment of the trajectories and societal transformations of regional societies, from their emergence to their ‘collapse’. Ultimately, the project has engendered new insights into long-term human-environment dynamics and in the process, underscored the pressing importance of acquiring data before this material legacy on the Earth’s surface is permanently lost to modern development.

View from the aircraft during lidar acquisition over mountainous terrain