Detection of many threatened aquatic mammals, such as manatees (Trichechus spp.), using traditional visual observation methods is associated with high uncertainty due to their low surfacing times, cryptic behaviors, and the environmental heterogeneity of their habitats. Rapid advancements in technology provide an opportunity to address these challenges. In this study, we aimed to quantify survey effort of unoccupied aerial vehicles (UAVs) for detecting the Vulnerable Amazonian manatee (T. inunguis). Using a closed population of manatees that is being rehabilitated within a lake at the Rainforest Awareness, Rescue, and Education Center in Iquitos, Peru, we calculated the number of repeat surveys needed to detect at least one individual with 95% (n = 3.10) and 99% (n = 4.76) confidence. We used both generalized linear mixed-effect models and Bayesian single-species and single-season detection models to determine the effects of the environment (water depth, water transparency, cloud cover, wind speed), time of day, and behavior (breathing, foraging, milling) on the time-to-detection and detection probability, respectively. Both models indicated a significant interaction between water depth and water transparency, causing an increase in the time-to-detection (β = 0.032; 95% CI = 0.028, 0.037) and a decrease in the probability of detecting manatees (α = -0.65; 95% CI = -1.3, -0.007), which was calculated to be 0.62 (95% CI = 0.23, 0.94). Due to the similarities between the lake and in situ habitats, the results of this study could be used to design in situ UAV survey protocols for Amazonian manatees or other difficult-to-detect freshwater aquatic mammals and to monitor ex situ animals pre-and post-release, which should ultimately contribute to a better understanding of their spatial ecology and facilitate data-driven conservation efforts.