Tuesday, August 4, 2009 - 8:20 AM

COS 17-2: Identifying bat echolocation calls at both species and individuals level: application of a blind source separation method

Lionel Humbert and François Fabianek. Université du Québec à Montréal

Background/Question/Methods

Researches on bat behavior often reach some limitations during survey such as individuals identification and species recognition. Radio tracking and telemetry are good tools to avoid these problems, however cost reduce the number of individual observed. Here, using echolocation signals we attempted to see if the independent component analysis, a blind source separation technique, could separate echolocation calls from different species and individuals. The goal was to evaluate the possibility to obtain a single signal for each individual, which can constitute its behavioral sequence. To test this method we used time-expanded echolocation signals from 10 individuals represented by height species found throughout Western Europe. We built artificially field recording conditions by mixing these signals together and added noise.

Results/Conclusions

The analysis was able to differentiate a single individual within noise when the original bat signal just represented 25% of the analyzed data. Moreover, we were able to recover the 10 individual signals separately inside a record of 18 mixed echolocation signals of 680 milliseconds. These first tests show the power of the analysis in acoustic signals recovering. We hope these finding will open new researches perspectives for study, survey and conservation of bats.