UAV-based deep learning and spatial analysis for mapping coastal plastic debris distribution

Abstract

Plastic waste accumulation is a major environmental challenge in enclosed and semi-enclosed coastal basins where restricted hydrodynamic exchange can promote long-term litter retention. This study assesses surface macroplastic litter along the Neftchala shoreline of the western Caspian Sea using UAV imagery, YOLOv4-based detection, GPS field validation, and spatial analysis. A total of 7,192 litter items were detected within approximately 0.385 km2, corresponding to an average density of 18,684 items/km2, while hotspot densities exceeded 60,000 items/km2. The distribution was strongly clustered, with a Nearest Neighbour Index of 0.49 and three shore-parallel accumulation zones. Detected litter was spatially associated with shoreline proximity, low-relief terrain, and reed-dominated vegetation, particularly desiccated reed belts exposed by recent Caspian Sea-level decline. Because the survey represents a single-date observation, these relationships are interpreted as spatial associations rather than direct causal mechanisms. The workflow supports targeted cleanup and monitoring across semi-enclosed coastal environments and broader evidence-based environmental management.