Survey bias in sampling François’s langur in the Karst mountains: An unmanned aerial vehicle (UAV) and accessibility perspective

Roadless areas in forested and rugged mountains pose unavoidable challenges to field surveys, often leading to wildlife sampling bias in terms of population, behaviour, and response to human interference. Using the endangered François's langur ( Trachypithecus francoisi ) as a model species, we aimed to compare the impacts of different survey methods, with varying levels of road accessibility on the evaluation of population size, the response of species to human disturbance, and the habitat distribution prediction by an ensemble species distribution model in the Dashahe National Nature Reserve, Guizhou, China. François's langurs live in some of the most inaccessible habitat among primates. To date, the best methods of survey have been interviews with locals and line transects along existing roadways. These methods may be biased by human presence and activities. Drones provide a means by which langur densities can be assessed across an inaccessible landscape, reducing researcher accessibility bias. We found that, compared to the interview survey, which have long been the only survey method in this region, the UAV survey significantly increased the species detection accessibility (SDA) of François's langur in roadless areas (population size estimation: interview survey recorded 37 individuals at mean SDA of 46.92 ± 21.06 km vs. UAV survey recorded 62 individuals at mean SDA of 57.96 ± 15.85 km). The unmanned aerial vehicle -methods vehicle (UAV) survey indicated that the occurrence of François's langurs decreased when the intensity of human modification exceeded 0.4, whereas the transect method showed a higher modification threshold that reached 0.6. This indicated that UAV survey exhibited a higher degree of sensitivity to human activities. Furthermore, while the suitable area identified by the UAV model (74.8 km²) was positively correlated with road accessibility ( p < 0.01), the results of the all-survey model (84 km²) showed a different pattern ( p < 0.001), with larger area identified as suitable habitat compared to those predicted using the other two survey methods. Our results suggest that, compared to other traditional surveys, UAVs increase the probability of detecting François's langurs in roadless areas (30.1%), eliminate overestimation of their tolerance to human disturbance, and help refine species distribution modelling. We recommend that UAV surveys be incorporated into routine wildlife surveys and conservation management studies to enhance our understanding of threatened species and reduce underestimation of human disturbance on these species.