Impact of deep learning and post-processing algorithms performances on biodiversity metrics assessed on videos

PLOS One | https://doi.org/10.1371/journal.pone.0327577 August 11, 2025 8 / 14 more than 10% to the species richness [18] and play important functions for ecosystems [38]. Hence, in most studies, processing rates should at least be 1 image per second to detect scarce species. Eventually, with all processing rates above 1 frame per second, all species were detected by detection models. Regard- less of the performance of the detection models, for processing rates of 0.25 and 0.5 image per second, only the rare species present < 1% of the time, were not detected. Hence, overall, the false negatives from the detection model (i.e., individual missed on images) did not markedly affect the estimated composition of the assemblage. It is important to note that although errors in detecting rare species at the level of individual frames may result in underestimating their presence, combining observations from multiple frames typically diminishes this bias when assessing broader ecological indicators. Consequently, despite the occurrence of detection errors at the frame level, their overall influence on biodiversity metrics can be reduced through repeated sampling. This partly accounts for our simulations indicating a general overestimation rather than underestimation of species richness. Fig 2. Effect of video processing rate and algorithms performance on the estimation of the abundance-structure of simulated communities. Estimation of abundance-structure as a function of video processing rate (fps, frame per second) for 45 automated analysis models resulting from 9

detection models, with their respective recall (“det_recall” – columns) and precision (“det_prec” – rows) performances and 5 classification models (accu- racy (“cla_acc”) as colors). Each dot represents the average over the 10 simulations with corresponding standard error as vertical bars. The Bray-curtis similarity measures the difference in species abundances estimated after automated processing using algorithms and species abundances actually visible on videos. https://doi.org/10.1371/journal.pone.0327577.g002