Non-invasive habitat evaluation through vibroscape monitoring

The structure and dynamics of the biological components of the vibroscape were analysed at different temporal scales, ranging from diel variations to seasonal changes. Substrate vibrations were recorded from plant stems using laser vibrometers or accelerometers. Initial analyses of vibroscape recordings were annotated manually, with vibrational events manually assigned to vibrational signal types (VST) based on their temporal and spectral characteristics. Later, we began using automatic detection and classification of vibrational signals based on deep audio embeddings and transfer learning to increase the speed of analyses. Our models show approximately 75% precision in detecting the most common signal types observed in our recordings. The highest abundance of VSTs was observed during the day at the beginning of July (middle of summer). The overlap of these signals in the time and frequency domains was significantly smaller than would occur just by chance. This reveals a partitioning of the vibrational communication channel used by many arthropods in the meadow. We propose that extracting information from vibroscape recordings could be an effective non-invasive habitat evaluation approach, encompassing a higher percentage of living beings than acoustic monitoring alone, thereby introducing the field of ecotremology.