Coastal paleo-landscape reconstruction in the Mediterranean context: from field data to AI tools

This PhD research develops an integrated framework for reconstructing Mediterranean coastal paleo- landscapes by combining geomorphological field investigations with explainable supervised machine learning techniques. The study focuses on the identification and spatial analysis of relict coastal landforms associated with Quaternary sea-level highstands, with particular emphasis on the Last Interglacial, the most recent interval during which global mean sea level exceeded present levels and widely regarded as a key analogue for future sea-level rise under warmer climatic conditions. The workflow was applied to two tectonically contrasting Mediterranean coastal systems: the Cilento Promontory (Southern Italy), a microtidal and tectonically quasi-stable margin, and the Laconian Peninsula (Southern Greece), an uplift-dominated forearc setting. The methodology integrates geomorphological mapping, GNSS surveys, UAV photogrammetry, and GIS-derived morphometric variables within a supervised Random Forest classifier. The training dataset was constructed using independently validated paleo–sea-level indicators derived from field surveys and previous literature and systematically stored in the PALEOScape geodatabase. These labelled indicators, together with associated predictor variables, were used to train the model, enabling continuous spatial mapping of coastal paleo-landforms—including marine terraces, paleo-sea cliffs, and wave-cut platforms—even in sectors where direct sea-level markers are absent. Model performance was assessed through statistical validation metrics and systematic spatial ground-truthing, confirming the robustness of the indicator-driven workflow across distinct geomorphological contexts. The comparison between these two end-member settings shows how Quaternary sea-level forcing may be preserved in different geomorphic expressions: in Cilento, MIS 5e is recorded as a polyphase highstand signal, whereas in Laconia the MIS 5a–5c–5e sequence forms a staircase of uplifted marine terraces. By integrating field-validated indicators, morphometric analysis, and explainable machine-learning classification, this research advances a scalable and objective framework for coastal paleo-landscape reconstruction and strengthens the quantitative interpretation of Mediterranean coastal evolution.