Automated mapping of dry stone walls in terraced landscapes using airborne laser scanning data

Abstract Context Dry stone walls are a prominent landscape feature integrating archaeological, ecological, and landform process aspects, but inventories of these features are often missing. Airborne laser scanning (ALS) provides data for efficient area-wide detection of linear landscape features like dry stone walls. Objectives We developed an automated approach to generate a comprehensive inventory of dry stone walls in terraced landscapes using statewide ALS data. Our objectives were to derive high-resolution digital terrain models (DTMs) and compare different methods to extract dry stone walls from DTMs and DTM-based features. Further, we produced a map of dry stone walls for the study site based on the best-performing method. Methods The study site is the 213 km 2 Wachau UNESCO World Heritage Site (Austria). We generated reference data for 206 ha for model training and testing. We investigated Random Forest classification of Canny edges and deep learning (DL) segmentation architectures on the DTM and DTM-based feature composites. Results Model evaluation yields F1 scores of 0.72 for the Random Forest classifier and 0.84–0.86 for the DL models. Comparable performance is observed when training on RGB composites of DTM-based features and on the normalized DTM alone. Convolutional and transformer-based architectures show no clear performance differences, indicating that DL-based detection of linear topographic features is robust to both input feature design and model architecture when using high-resolution elevation data. The area-wide application results in a final dataset of 925 km of detected dry stone walls across the Wachau region. The derived density map shows dry stone wall densities up to 2740 m/ha. Conclusions Our results show that publicly available ALS data allow reliable, automated mapping of dry stone walls across terraced landscapes. The Wachau-wide mapping provides a basis for heritage documentation, habitat assessments, and future landscape-scale monitoring of the state of dry stone walls. The study demonstrates how DL methods can enhance landform mapping for landscape management and monitoring.