The ability of an animal to detect environmental cues is crucial for its survival and fitness. In bats, sound certainly plays a significant role in the search for food, spatial navigation, and social communication. Yet, the efficiency of a bat’s echolocation could be limited by atmospheric attenuation and background clutter. In this context, sound can be complemented by other sensory modalities, like smell or vision. Spix’s disc-winged bat (Thyroptera tricolor) uses acoustic cues from other group members to locate the roost (tubular unfurled leaves of plants in the order Zingiberales). Our research focused on how individuals find a roost that has not been yet occupied, considering the urge to find a suitable leaf approximately every day, during nighttime or in daylight. We observed the process of roost finding in T. tricolor in a flight cage, manipulating the audio/visual sensory input available for each trial. A broadband noise was broadcast in order to mask echolocation, while experiments conducted at night significantly reduced the amount of light. We measured the time needed to locate the roost under these different conditions. Results show that with limited visual and acoustic cues, search time increases significantly. In contrast, bats seemed capable of using acoustic and visual cues in a similarly efficient manner, since roost search showed no strong differences in duration when bats could use only sound, only vision, or both senses at the same time. Our results show that non-acoustic inputs can still be an important source of information for finding critical resources in bats.

We captured individuals of Thyroptera tricolor in a secondary forest in southwestern Costa Rica. We conducted experiments in a flight cage (2.5 x 3.5 x 5.5 m) containing a roosting resource available for the bat (a furled leaf of Heliconia sp. or Calathea luthea). In daylight experiments, the flight cage was made of saran shade cloth while in night experiments it was made of a double-walled cloth to reduce the amount of light. Two Ultrasonic Omnidirectional Dynamic Speakers were positioned on both sides of the leaf, at 1 m distance. We tested the time needed to enter the roost under different treatments, that we named after the sensory cue available for the bats: "Vision" (a playback broadcasted a broadband noise masking the frequencies of Thyroptera's echolocation, with daylight), "Sound and Vision" (no playbacks, with daylight), "Lessen input" (broadband noise, during nighttime), "Sound" (no playbacks, during nighttime). "Noise Control" was an additional treatment in which we broadcasted a sound not masking echolocation frequencies. We used that test as a control to check if bats' behaviour during the playback of the broadband noise was a response to a noisy environment or to effective acoustic masking. In the .csv file, you can see the times needed for each individual to reach the roost, in seconds, under each treatment. Maximum duration for each test was set at 300s.

The data were then processed using a Bayesian regression to evaluate the effect of the different sensory inputs (categorical predictor) in the time required for the bat to enter the roost (response, modeled with a lognormal function), including individual as random effect (varying intercept). The regression models included the observations for individuals that were tested more than once in the same experimental condition. Regressions were run in Stan (Stan Development Team 2021) through the R package brms (Bürkner 2017; R Core Team 2021). Effect sizes are presented as median posterior estimates and 95% credibility intervals as the highest posterior density interval. We compared the model with an intercept-only model (null model) using the Bayesian leave-one-out information criterion (LOOIC, Vehtari et al. 2017) with the R package loo (Vehtari et al. 2020). We conducted multiple comparisons of sensory input treatments (similar to post hoc tests in frequentist statistics) using the joint posterior distribution of the model parameters with the function hypothesis from the package brms (Bürkner 2017). Models were run on four chains for 5,000 iterations, following a warm-up of 2,500 iterations. The effective sample size was kept above 3,000 for all parameters. Performance was checked visually by plotting the trace and distribution of posterior estimates for all chains. We also plotted the autocorrelation of successive sampled values to evaluate the independence of posterior samples. A potential scale reduction factor was used to assess model convergence and kept below 1.05 for all parameter estimates.

Microsoft Excel or Libre Office Calc; Google Chrome or any other browser