Quantifying airborne vessel noise on large rivers: A novel approach combining acoustic drifters and machine learning

Vessel noise in large rivers poses a growing threat to riparian ecosystems and human communities, yet its large-scale airborne distribution remains poorly monitored. This study introduces an innovative framework combining mobile observations with machine learning. We first conducted extensive measurements of surface airborne noise along the Yangtze River using self-developed Acoustic Drifters. This mobile sensing approach captures the fine-grained spatial heterogeneity of noise along a critical transport artery, overcoming the limitations of static monitoring. Based on the collected data and vessel information, we constructed a Graph Neural Network model to dynamically predict vessel noise. The model demonstrated strong robustness, confirming vessel activity is the primary noise source. Using this model, we successfully generated high-resolution, near-real-time noise maps, revealing chronic exposure to high noise levels near shipping lanes to levels known to impair wildlife foraging and survival. This quantitative assessment directly links mobile traffic sources to their spatially-explicit acoustic footprints. This provides a spatially-explicit evidence base for assessing the cumulative environmental and health risks of shipping noise. Ultimately, our transferable methodology offers a crucial tool for developing global environmental policies to mitigate the acoustic impacts of waterway transportation.