Seven frog species (Crinia parinsignifera, C. signifera, Litoria peronii, Limnodynastes tasmaniensis, L. dumerili, L. fletcheri and Neobatrachus sudelli) were subject to electronic audio monitoring using 20 call recorders spread across the Koondrook-Perricoota Forest. Two methods for identifying calls of the seven species from the audio data were employed with the results being subsequently considered in terms of their ability to monitor frog breeding responses to environmental conditions. A manual extraction method undertaken during specific periods of wetland inundation produced high detection rates, whereas an audio recognition software approach produced lower detection rates overall but produced data over a longer period. While it appeared that the manual method has potential to produce more accurate results than the audio recognition software method, implementation of three recommendations stemming from the analysis of data produced by this study could greatly increase the efficacy of monitoring methods that employ software based analysis of collected audio data. These recommendations are to (1) produce many spatially varied representative call recognition models for the software recognition matching algorithm; (2) if using the raw output of the recogniser software without validating the data, identification of 11 candidate calls would result in at least a 95% chance that at least one true positive call was observed for most species (L. fletcheri requires 18); and (3) increasing call recorder frequency to hourly rather than daily 5-minute periods would likely vastly increase probability of call detection using call recognition software.
|Publisher||Charles Sturt University|
|Commissioning body||Forestry Corporation of New South Wales|
|Number of pages||19|
|Publication status||Published - 23 Nov 2018|