An Open-Source Machine Learning–Based Methodological Approach for Processing High-Resolution UAS LiDAR Data in Archaeological Contexts: A Case Study from Epirus, Greece

Abstract This study shows and discusses an innovative approach devised for archaeological feature detection using unmanned aerial system (UAS) LiDAR and an open-source probabilistic machine learning framework. The methodology employs a Random Forest classification algorithm within CloudCompare’s 3DMASC plugin to analyse dense LiDAR point clouds. The main steps include classifier training, hyperparameter adjustment and point cloud segmentation to produce digital terrain models (DTM), digital feature models (DFM) and digital surface models (DSM). Experimenting different parameters led to the determination of the best set to be employed for the training model. Subsequent data enhancement with the Relief Visualisation Toolbox (RVT) refines the visibility of archaeological features, particularly within complex and heavily vegetated terrain. The use case selected to validate this approach is the site of Kastrí-Pandosia in Epirus (Greece), which is particularly suitable for LiDAR analysis by UAS. This approach significantly improves archaeological detection and interpretation, revealing previously inaccessible or obscured microtopographic and structural features. The results highlight the site’s defensive walls, terracing and potential anthropogenic routes, underlining the methodology’s effectiveness in detecting archaeological landscapes at multiple levels. This study emphasises the utility of accessible and open-source solutions for the identification of archaeological features, promoting cost-effective methods to improve the documentation of sites in remote or difficult locations.