Paleo-seascape reconstructions along the Cilento coasts (Tyrrhenian sea) by innovative AI approach

This study wants to evaluate the paleo-landscape evolution since the MIS 5 of Cilento coastal sector (Tyrrhenian Sea) by means of an innovative methodology that combines traditional geostatistical techniques with machine learning applied to paleo sea-level markers (SLMs), including SLIPs, TLPs, and MLPs, integrated with morphometric DTM analysis.

The study area is the Cilento coastal sector located along the Tyrrhenian coast of Southern Italy. This area preserves several features witnessing sea-level fluctuations but, despite the area having been intensely studied since 1940, no comprehensive database encompassing the local paleo sea-level evidence existed prior to this research. The area has been chosen due to its tectonic stability, which ensures consistent correlations to the same age of the markers having the same altimetric position.

The initial phase involved collecting and updating data from the Campanian Natural Cavities Inventory (Russo et al., 2005) and conducting extensive field surveys. This effort resulted in the PALEOScape geodatabase (Sorrentino et al., 2023), where data were classified using new indexes: the Environmental Index Point (EIP) and Environmental Limiting Point (ELP), reflecting varying levels of uncertainty in paleo-shoreline positions. Subsequently, morphometric DTM analysis and spatial queries overlaid the geodatabase, generating a Random Forest dataset trained on 500 records. This methodology was validated with two levels of validation: a first level of statistical analysis on the training dataset, and a second level across four deeply surveyed coastal areas serving as ground truth. The accuracy of the resulting models ranged from 0.7 to 1. In this way, a prediction value was obtained even where no markers are present.

The approach enabled extensive paleoenvironmental reconstructions for the Cilento region during the Last Interglacial Period, modelling scenarios at both high and low sea-level stands. These findings provide an evolutionary model of coastal changes, demonstrating the utility of integrating traditional and advanced techniques for robust paleoenvironmental analysis.

In conclusion, this work offers a novel framework for large-scale paleo-coastal reconstructions, facilitating the evaluation of shoreline evolution and its implications for past human activities and future coastal management.