Optimizing small mammal surveys in Neotropical fragmented landscapes while accounting for potential sampling bias

Maximizing cost-efficiency of biological surveys is of great importance to assess threats and monitor changes in tropical landscapes, particularly when survey costs are high as in Neotropical small mammals. Species detectability is mostly affected by local abundance but can further vary according to the forest size in fragmented landscapes. For example, species detection in smaller fragments may be higher due to either spatial constrains or forest vertical compression. Here, trade-offs between duration of trapping sessions and the cost of obtaining accurate small mammal individual and species data are investigated across a range of forest sizes, while accounting for potential sampling bias. Using a combination of live (Sherman and wire mesh) and pitfall traps, patterns of cumulative capture-recapture rates and community composition were examined across 16 night trapping sessions in 42 transects nested within 14 insular forest fragments of different sizes, and two continuous forest sites in the Central Brazilian Amazon. Based on 40,254 trap-nights, 604 captures were recorded from 22 identifiable species. Cumulative number of captures gradually increased along the entire trapping session, and on average, nearly half of all species required more than eight trapping nights to be detected at each sampling transect. Sampling bias was responsible for increased cumulative recapture rates in smaller fragments, but did not affect cumulative capture rates, or time to first detection of each species. Instead, time to first detection was mostly positively affected by species abundance. Our results highlight the higher effectiveness of long trapping sessions (i.e. > 10 nights) to detect uncommon species, increase sample sizes and recapture rates. Overall, if small mammal species abundance is high, sampling intensity can be reduced. However, regardless of species abundance, large fragments and continuous forest sites should be more intensively sampled to reduce any potential sampling bias.