Methodological choices influence ecological inference in passive acoustic monitoring of a Neotropical nightjar

Liliana Piatti, Daiene Louveira Hokama de Sousa, Beatriz dos Santos Oliveira, Alyson Vieira de Melo, Larissa Sayuri Moreira Sugai, Diogo Borges Provete 

Passive acoustic monitoring is increasingly used to investigate species activity and habitat use through occupancy analyses. Yet, the complex analytical workflow, from automated detector choice to confidence thresholds and statistical modeling framework, is amongst the factors that influence ecological inference, and the extent these decisions affect modelling outputs is poorly debated. Here, we use acoustic data from a Neotropical nightjar (Nyctidromus albicollis) in the Pantanal wetlands, Brazil, to evaluate how these decisions lead to different conclusions about the environmental drivers of nocturnal vocal activity. We processed 97,906 minutes of audio files from seven sites using two classification algorithms (a locally trained custom model and a pre-built global model), each evaluated at two confidence thresholds based on precision and F1 metrics, yielding four detection datasets. We modeled vocal activity in relation to temperature, relative humidity, and lunar illumination using both conventional regression and Bayesian occupancy models that explicitly separate the contributions of ecological and observation processes. All models agreed on the direction of environmental effects: vocal activity declined with increasing temperature and humidity and was weakly associated with lunar illumination. However, the magnitude of the effects differed substantially across detector–threshold combinations. Detections based on the prebuilt model overestimated the effects of all variables, with predicted occupancy at low temperatures ranging from over 90% (prebuilt) to below 25% (custom). High-precision thresholds were far better calibrated, highlighting the importance of prioritizing precision for more reliable inference under hierarchical modeling. The naïve regression approach produced attenuated effect sizes and narrower uncertainty intervals compared to the occupancy framework, with discrepancies depending on the detector used. Our results demonstrate how methodological decisions across the analytical workflow shape contrasting ecological interpretations. We recommend that ecological inference using acoustic monitoring explicitly address the impact of different decisions during the analytical process and incorporate detection uncertainty through hierarchical modeling.