Quantifying the effect of forest edge on tropical fauna using explainable ecoacoustics metrics
Tropical forests are biodiversity hotspots but increasingly fragmented by human activities, multiplying forest edges that alter microclimates and species dynamics. While edge effect on plants is well documented, its impact on fauna remains less ex-plored. Ecoacoustics provides a non-invasive and passive approach to monitor faunal communities, yet the complexity of soundscapes and the opacity of machine learning tools often limit ecological interpretability. Here, we show that acoustic diversity in New Caledonian ultramafic forests exhibits a detectable edge effect, with sound-scapes converging towards homogeneity starting 100 m from the forest edge. Using passive acoustic monitoring and 59 ecoacoustic indices, we combined machine learn-ing, Representational Similarity Analysis, and explainable statistical methods from psychophysics. Results reveal a significant correlation between distance to edge and acoustic diversity, particularly pronounced during day-time. Indices such as NDSI, Hf, and SKEWf contributed most to detecting the effect, whereas some other in-dices added noise rather than informative signal. These findings match the botanical threshold described by Blanchard et al. (2023) and highlight both the promise and limitations of indice-based ecoacoustics. By integrating explicability methods, we offer a transparent framework to disentangle complex ecological signals, advancing ecoacoustics as a pertinent tool for studying forest fragmentation and guiding biodi-versity conservation.