Improving primate detection using thermal infrared imaging: availability and observer errors in drone surveys of mixed tropical forests

Drones with thermal infrared (TIR) cameras are increasingly used to survey terrestrial endotherms, including primates, but the relative lack of standardized protocols hampers their effectiveness. Detection errors, both observer (animal is available for detection but missed/misidentified by the observer) and availability errors (animal is present but unavailable for detection), vary across species, habitats and flight parameters, and may be exacerbated by anti-predator responses. We evaluated the suitability of drone-mounted TIR cameras in surveys of long-tailed macaques (Macaca fascicularis) by measuring detection errors for different flight altitudes (70, 60, 50 and 40 m above ground level [AGL]) and camera angles (45° and 90°) in two secondary forest habitats in Mauritius, characterized by open canopy with low thermal contrast and closed canopy with high thermal contrast. First, we determined the true size of three habituated macaque groups with full-day follows, and assessed their response to the drone. Next, we measured availability error (detection rate) in TIR videos after correcting for observer error (misidentification rate) using corresponding red-green-blue (RGB) videos. Finally, we assessed whether adjusting counts with mean detection rates improved group size estimates for different flight parameter combinations. Both error types were lowest for morning flights at 50 and 60 m AGL (40 and 50 m above canopy respectively) and with 45° camera angle in high thermal contrast conditions. The higher detection rates at 50 and 60 m compared to 40 m resulted from milder anti-predator behaviours (increased vigilance instead of dispersal). In secondary forest habitats with few dominant canopy species (e.g. Eucalyptus, Litsea and Pinus spp.), and for primate species with behavioural ecology comparable to long-tailed macaques, we recommend surveying at 40–50 m above closed-canopy vegetation with a 45° camera angle. With other flight parameters, we recommend incorporating detection rates from this study, along with their errors, into final counts.