Quantifying ancient architecture: Stone-level machine learning reveals architectural variation and settlement growth at Shivta.

Machine learning on drone and close-range imagery at the Late Antique settlement of Shivta digitized ~235,000 surface stones and 12,000 stones in 604 walls, revealing masonry variation linked to settlement growth.

• Why it matters: Quantifying architectural variability has been labor-intensive, structure-limited, or reliant on coarse volumetric estimates, limiting settlement-wide comparisons of construction practices over time.
• What they did: The framework digitized individual stones and quantified stone size, shape regularity, and masonry arrangement across walls, households, insulae, and the settlement using two complementary image datasets.
• The result: Higher masonry scores cluster around churches and the settlement boundary, and masonry variation maps an earlier boundary and supports outward expansion from a southern core, with larger, more regular stones used during Shivta’s principal growth phase.