Articles by Gabriele Mazzacca

This paper reports preliminary outcomes of the 'Hidden Landscapes of Latium project', applied to the Presidential Estate of Castelporziano, an exceptionally well‑preserved coastal landscape in the central Mediterranean and, concurrently, a highly challenging forested context for landscape archaeology. The study area, about 60 km², characterized by dense and continuous vegetation cover, was surveyed by airborne LiDAR acquisition at very high point density (in average 350 pts/m²), with the goal to facilitate under‑canopy archaeological detection in a highly challenging Mediterranean environments. We provide a critical appraisal of the acquisition strategy, with particular attention to sampling density and flight design as determinative factors for the detection of subtle anthropogenic micro‑topographies. The acquired dataset was subjected to a semantic‑segmentation processing pipeline that integrates multi‑level and multi‑resolution machine‑learning techniques for the automated classification of ground, vegetation, and structural elements. Results indicate that the combination of archaeologist-addressed, high‑resolution LiDAR acquisition and advanced semantic classification substantially enhance the interpretability of forested archaeological landscapes and enables the identification of features previously obscured by vegetation. Castelporziano thus constitutes a valuable testbed for evaluating the applicability, transferability, and limitations of LiDAR‑based methodologies within Mediterranean landscape archaeology.

This paper presents an integrated geophysical and UAV-LiDAR investigation of the Rusellae (Roselle) hinterland, carried out within the Emptyscapes framework aimed at moving beyond site-based approaches toward a continuous, landscape-scale reconstruction of settlement and land-use dynamics. While extensive field survey, aerial imagery and large-scale magnetometry are effective in the open agricultural portions of the Ager Rusellanus, densely vegetated hill slopes remain a major source of knowledge bias. To address this gap, we combined (i) new magnetic prospections (ca. 70 ha) in the south-western sector of the study area with (ii) a drone-based high-density LiDAR survey (ca. 550 ha) acquired with a RIEGL miniVUX-3 at low altitude and controlled speed, achieving an average density >700-800 pts/m². Point clouds were processed through a multi-level, multi-resolution semantic segmentation workflow, using a Point Transformer deep-learning architecture to classify ground, vegetation and above-ground structures and to produce high-resolution DTMs and DFMs for GIS-based analysis. Results reveal a markedly more structured landscape in wooded areas than previously documented, including extensive terracing systems, mobility corridors, funerary districts with rock-cut hypogea, and traces of later productive activities. The study highlights both the potential and the limits of high-density UAV-LiDAR and ML-based processing, stressing the need for targeted ground validation in Mediterranean under-canopy contexts.