Physically-based modelling for retrospective detection of archaeological proxies (cropmarks)

Area of Science:

• Archaeological prospection
• Remote sensing
• Geophysics

Background:

• Spectral anomalies like cropmarks are key to finding hidden archaeological sites.
• Understanding cropmarks requires analyzing vegetation stress and radiative transfer dynamics.

Purpose of the Study:

• To develop a physically based modeling strategy for interpreting cropmarks as vegetation stress.
• To create a reproducible pipeline merging physical simulation with machine learning for archaeological detection.
• To address data scarcity in heritage research through synthetic data generation.

Main Methods:

• Utilized PROSAIL model in forward and inverse modes for radiative transfer dynamics.
• Generated synthetic spectral datasets to augment limited field observations.
• Applied an ensemble of machine learning algorithms trained on synthetic data for classification.

Main Results:

• Achieved over 90% detection rates on historical observations, proving retrospective application.
• Identified plant growth phase, particularly peak greenness, as crucial for performance.
• Demonstrated that injected noise modestly improves model robustness.

Conclusions:

• Established a reproducible pipeline for archaeological prospection using physical simulation and machine learning.
• Synthetic data generation is a viable solution for data scarcity in heritage research.
• The developed models can be applied to archive aerial and satellite imagery, transforming remote sensing in heritage research.