Estimating abundance of aggregated populations with drones while accounting for multiple sources of errors: A case study on the mass nesting of Giant South American River Turtles

Abstract Counting animals when populations are spatially aggregated (e.g., breeding or nesting colonies, stopover or haul‐out sites) enhances the accuracy and efficiency of survey efforts for abundance estimation. Orthomosaics generated from drone images are commonly used to count aggregated populations, but these counts are subject to detection errors that are often overlooked in abundance estimation. Motivated by the need for a monitoring protocol for mass nesting events of Giant South American River Turtles ( Podocnemis expansa ), we develop a novel modelling approach to estimate the abundance of spatially aggregated wildlife populations using drone‐based counts in orthomosaics while accounting for multiple sources of error. We use a combination of mark‐resight data and overall population counts to account for: (i) open population during the nesting event; (ii) individuals unavailable for detection during flight; (iii) double counts due to the orthomosaic building process; (iv) marked individuals detected in the mosaic with unidentifiable marks. From the mark‐resight data, we estimated that the daily nesting probability is 0.37, and that 35% of the individuals that used the sandbank during the night are present during the morning drone flight. We also found that 20% of the turtles walking in the orthomosaic are double counts, and that the probability of identifying the mark in the carapace is 0.78. The total population size was estimated as ~41,000 turtles for the 12 days of nesting season, marking the current world's largest known aggregation of freshwater turtles. By comparing our approach with an abundance estimate based on a simpler model and with visual ground counts, we demonstrate the benefit of our approach and the importance of accounting for the multiple sources of error when counting animals in orthomosaics. Synthesis and applications . The developed approach can be applied to several contexts to efficiently survey spatially aggregated populations using drone‐derived orthomosaics, and to understand phenology at these aggregation sites. We provide general recommendations for planning surveys and discuss implementations of our approach using other types of marking methods and model assumptions.