batIdent, the successor of bcDiscriminator, extracts species information from call measurements derived by bcAdmin in less than a second using an objective, statistical method. Currently for european bat species.
One of the biggest problems of using acoustical methods for bat activity acquisition is the discrimination of species by their calls. While some genera (like Pipistrellus) are more or less simple to determine on a secure level, for others this is close to impossible without elaborate methods (like Myotis). Up until now one has to take measurements manually and compare these with numbers from few publications, time-consuming and non-satisfying. Quite often the awarding authority is not willing to pay this sumptuous task completely.
Due to differing expertise on call analysis and bioacoustic knowledge we still face low acceptance and reservation about acoustical monitoring and species discrimination. Our solution - batIdent - offers you objective species recognition on a per call base and is freely available as open-source software. It's the only available application world-wide that does this so far. The application uses call measurements we have derived in the last few years. It harnesses the power of modern statistical approaches (randomForest and support vector machines) to reliably discriminate species upon their calls. Results for up to three species are saved to file and can be imported into bcAdmin. Updates of the underlying algorithms can be installed easily, thus adding new species or a better data base for discrimination is possible.
The following scheme displays the identification tree and its steps as it is used in batIdent:
While using about 600 trainings calls per species, nevertheless confusions of species occur. Due to overlapping call repertoires this must be expected. The following graph supplied by NycNoc GmbH displays the out of bag errors, which were recorded with a single call analysis using randomForest: