Road disturbance drives a more simplified soundscape in temperate forests revealed by deep learning and acoustics indices

Global increases in road networks have been a significant ecological stressor. There is a growing awareness of its negative effects on wildlife and soundscapes, particularly through habitat fragmentation and the introduction of anthropogenic noise. Passive acoustic monitoring (PAM) of biodiversity may provide integrative indices for assessing road effects. However, little is known about the responses of soundscape characteristics to roads with different traffic volume levels, especially in temperate forests. Here, we combined acoustic scene classification (ASC) and acoustic indices to examine the effects of road disturbance on soundscape attributes and composition in Northeast China. We collected and analysed over 3300 h of recordings from 28 sites across high-, medium-, and low-traffic roads. We classified each recording into different acoustic scenes based on the ResNet50 Convolutional Neural Network (CNN) and calculated the acoustic complexity index (ACI), bioacoustic index (BIO), acoustic diversity index (ADI), normalized difference soundscape index (NDSI), and power spectral density (PSD) values. Our results showed that acoustic indices and the daily audibility of birds, insects, amphibians and mammals significantly responded to road disturbances over 24 h diel cycles but with contrasting patterns; roads with intensifying traffic volume advanced the peak of biophony (BIO). As the traffic volume increased, the soundscape composition became more simplified, as measured by significantly lower ADI2K and NDSI values and significantly fewer daily audibility of the mammals. The ASC results further revealed that vehicle and insect sounds dominated the soundscape along the high-traffic road where nocturnal insects became the strongest acoustic markers. This study highlights the complex interactions between road disturbances and diverse biological taxa and presents important evidence that the effects of traffic noise and road networks should be considered in ecosystem conservation and management plans for the wildlife and acoustic communities of the landscape. We argue that the ASC combined with multiple acoustic indices is required to adequately account for the complex responses of soundscapes to large-scale anthropogenic disturbances.