Warra Tall Eucalypt Acoustic Data Collection - TERN 

This dataset contains audio files from Warra Tall Eucalypt SuperSite. Warra Tall Eucalypt SuperSite was established in 2012 and is located in a stand of tall, mixed-aged Eucalyptus obliqua forest (1.5, 125 and >250 years-old) with a rainforest / wet sclerophyll understorey and a dense man-fern (Dicksonia antarctica) ground-layer. The site experienced a fire in January 2019, which consumed the ground layer and killed a high proportion of the understorey trees, but stimulated dense seedling regeneration. For additional site information, see Warra Tall Eucalypt SuperSite In 2012 an acoustic recorder was set up to collect audio data for a total of 12 hours per day, split between six hours around dawn and six hours around dusk. The recording schedule aimed at capturing morning and evening bird choruses while minimizing memory and battery requirements. A long-term spectrogram has been generated for each audio file to aid in data exploration. The sensor also recorded temperature, minimum- maximum- and mean-sound pressure levels. The sensor stopped working in 2019 when it was destroyed by a fire. Acoustic indices and false colour spectrograms were created for the recordings. Acoustic indices are summaries of the distribution of the acoustic energy in a recording. They are particularly useful for the analysis of long-term recordings of the environment and can be used to identify sound sources of interest, characterise the soundscape, aid in the assessment of fauna biodiversity, monitor temporal trends and track environmental changes. False colour spectrograms are visual representation of individual acoustic indices or combination of multiple indices. They can highlight the presence of specific sound sources, e.g. birds, insects or weather events, providing a tool for navigating long-term recordings. Data are made available through the data link. For downloading large amount of data, please follow these instructions How to download TERN's acoustic data in bulk 

We at TERN acknowledge the Traditional Owners and Custodians throughout Australia, New Zealand and all nations. We honour their profound connections to land, water, biodiversity and culture and pay our respects to their Elders past, present and emerging. The research site was established in 1995 by Forestry Tasmania and joined the TERN SuperSite Network in 2013. This work was funded by the Terrestrial Ecosystem Research Network (TERN), an Australian Government National Collaborative Research Infrastructure Strategy (NCRIS) project. 

Long-term acoustic recordings are collected to characterise the acoustic sources in the ecosystem. Recordings can be used to estimate biodiversity, monitor temporal changes in the soundscape, compare the acoustic characteristics of different locations, and assess the effect of particular events such as bushfires and floods. 

An acoustic sensor was set up to collect audio data as part of a continent wide long term monitoring project. The sensor was a Wildlife Acoustics Song Meter 2 equipped with two microphones. According to manufacturer's specifications the microphones sensitivity was -36±4 dB (0 dB=1 V/Pa at 1 kHz). The sensor was mounted on a star picket. Data were recorded for a total of 12 hours per day, split between six hours around dawn and six hours around dusk. Recordings were made as dual channel, three-hour long wac files, and were later converted into flac format. They had a sampling rate of 22,050 or 44,100 Hz and a depth of 16 bits. Long-term spectrograms have been created for the audio files and are avaialble through the data link. The sensor also recorded 'ancillary data' such as temperature, minimum- maximum- and mean-sound pressure levels. Acoustic indices were calculated using the software AnalysisProgram version 19.2.2.1. Long term recordings were resampled at 22.05 kHz and divided into one minute long segments. Acoustic indices were calculated for each segment. Spectral indices were calculated on one minute long spectrograms. Each spectrogram was created by first dividing the one minute long recording into frames of 512 samples each and then calculating the Fast Fourier Transform for each frame. The frequency resolution was 43.1 Hz. The spectra were smoothed using a moving average window of width three and spectral amplitude values are converted into spectral power or decibels (dB). Summary indices were calculated on the waveform envelope or were derived from the spectrograms. The wave envelope was created by taking the maximum absolute value in each frame. Absolute values were converted to dB. False colour spectrograms were produced for each spectral index and combination of indices. Please note the following points:

• dB values are in reference to a hypothetical signal of unit amplitude.
• Before the indices were calculated, noise had been removed form both the waveform and the spectrogram using a modified versions of Lamel's adaptive level equalization (Lamel et al., 1981).
• Entropy values were subtracted from 1 to obtain a measure of energy concentration which provides a more intuitive index.
• To create false colour spectrograms, indices were normalised between a minimum and maximum value. These values affect only the visualization and are provided in the configuration file in the data access link.
• The dataset contains variables labelled DIFsp and SUMsp which are used only for internal checks. They do not correspond to acoustic indices.