Passive acoustic monitoring (PAM) of the endangered African penguin (Spheniscus demersus) in the Stony Point Nature Reserve

The African penguin is a threatened seabird of the southern African coast, despite the status of a protected species the population is decreasing. My study aimed to assess the presence of active nests in a wild colony of African penguins using a non-invasive passive acoustic monitoring technique. The acoustic recordings were collected from February to April 2024 at the African penguin colony of Stony Point (South Africa) along with a visual census of the active nests. The acoustic recordings were collected using 10 Autonomous Recording Units (ARUs): Song Meter Micro (Wildlife Inc.) with a sampling rate of 48000 Hz that recorded every day from 4:00 to 7:00, and from 18:00 to 22:00 for 11 weeks continuously. The second part of data collection was the visual census of the active nests, a common practice used in African penguin monitoring. The census was operated in the same quadrant where the Song Meter was present. In the analysis, I used the Vocal Activity Rate index (VAR) which represents the number of calls per recording time. This index was used to quantify the Ecstatic Display Songs (EDS) of the penguins. To identify the EDSs, I used a custom-built Convolutional Neural Network (CNN) build for previous works. To validate the CNN performance, I compared the automated detections with a subset of recordings manually annotated using Praat, a software for audio analysis. In the analysis of active nests detection, I performed a glmmTMB (Template Model Builder) model to investigate if the VAR index is effective in the count of active nests in a wild colony of penguins. I continued the analysis with the development of a Full model and a Null model that was compared with a Chi-squared test. The model was run using the VAR as the response variable, and as fixed factors were considered the number of active nests counted during the census and the different types of habitats of the colony. The random factors operated in the model were the identification number of the ARUs (ID) indicating the different sampling sites and the week of recording. In the comparison between the Null and the Full model, I found a significant difference between the two, indicating that the fixed factors influence the response variable. The study compared the results of the visual census of active nests and the Vocal Activity Rate estimated per every sampling point. The statistical analysis evidenced a proportional relationship between the VAR index and the number of active nests. My thesis evidenced the important role of PAM in the detection of active nests in dense vegetation sites where the visual census can be hindered. That confirms how PAM is becoming an increasingly important component of the traditional survey. The encouraging results of the thesis confirm the importance of further applications of this technique in long-term monitoring plans of the population trend in African penguins.