Quantifying and mitigating recorder-induced variability in ecological acoustic indices

“This difference results in a negative correlation between AEI and ADI indices, as demonstrated by (Peet, 1974).”

“Many EAI are based on the Fast Fourier Transform (FFT) of the signal.”

“The advantage of using K-S distance is that it is sensitive to both the shape and position of the probability distribution, providing a clear comparison between distributions (James, 2006).”

“Moreover, by making EAI comparable across studies , we can use them in new applications such as machine learning and deep learning, which require features in the same space, i.e., without considering the device with which they were acquired (Purwins et al., 2019).”

“In this study, we address the challenge presented by this open issue, which mainly affects the growth of acoustic monitoring and is noticeably prominent in large-scale studies (Gibb et al., 2019).”

“The Entropy Index (H) was proposed by Sueur et al. (2008) as an indicator of heterogeneity and is based on estimating the Shannon uniformity over time and frequency.”

“EAI encode complex and biologically relevant information in a single numerical value, efficiently processing large volumes of data (Sueur et al., 2014; Eldridge et al., 2016).”

“Ecological acoustic indices (EAI) are one of the most common metrics to study soundscapes as they are easy to calculate and are assumed to have biological interpretability (Buxton et al., 2018; Sueur and Farina, 2015; Farina and Gage, 2017).”

“To validate our approach, we compute the eight EAI’s probability density functions (PDF) at each site using the Parzen method Robert (1976).”

“Due to their flexibility, EAI have been used for different aims, such as acoustic niches, uniformity, soundscape degradation, and environmental studies (Towsey et al., 2014; Shaw et al., 2021; Benocci et al., 2020).”

“The Acoustic Diversity Index (ADI) was proposed by Villanueva-Rivera et al. (2011) to determine the occupancy degree of acoustic niches.”

“However, several studies using the same indices provide contradictory re-sults by correlating with biological variables, generating doubts about EAI effectiveness, despite the prevalence of their use Bradfer-Lawrence et al. (2019a).”

“This problem will increase since nowadays researchers commonly have in their stocks high and low cost devices (SongMeter and BAR and Au-dioMoth) (Hill et al., 2019).”

“However, some species that make calls at low frequencies can generate values of NDSI 0 .”

“However, since these indices may provide similar information, using both indices may sometimes be redundant (Eldridge et al., 2018).”

“Step-1: Signal resampling Sampling frequency differences are often disregarded, Adams et al. (2012); Rempel et al. (2013).”

“Subsequently, traditional peak detection algorithms (such as proposed by Dumpala et al. (1982) or Sueur (2018)) are used to find prominent peaks.”

“The Number of Peaks index (NP) was proposed by Gasc et al. (2013) as an indicator of acoustic complexity using the number of prominent peaks in frequency (see Figure B.10).”

“Additionally, the AEI index has been used to differentiate acoustic activity at different times of the day and in different habitats Dr¨oge et al. (2021).”

“The intra-study variability of EAI has been attributed to various factors, such as the sensitivity to changes in environmental conditions, noise, and the sampling protocol Bradfer-Lawrence et al. (2019a); Metcalf et al. (2021a); Lillis and Mooney (2022); Zwerts et al. (2022).”

“However, this strategy does not consider differences between devices for frequencies above 500 Hz, so the variability due as frequency response persists Pieretti et al. (2015).”

“In this study, all indices were calculated using the Python scikit-maad library proposed by Ulloa et al. (2021).”

“Step-1: Signal resampling Sampling frequency differences are often disregarded, Adams et al. (2012); Rempel et al. (2013).”

“…although different works aim to identify the possible causes of this variability and propose alternatives to measure the effectiveness of EAI for characterizing various habitats, a straightforward standard implementation is needed (Ross et al., 2021; Zhao et al., 2019a; Jorge et al., 2018).”

“…although different works aim to identify the possible causes of this variability and propose alternatives to measure the effectiveness of EAI for characterizing various habitats, a straightforward standard implementation is needed (Ross et al., 2021; Zhao et al., 2019a; Jorge et al., 2018).”

“These indices quantify acoustic variability at the sampling sites, provide different information about the soundscape, and their use spans different habitats and environmental scales Zhao et al. (2019b).”

“As gleaned from the literature, intra-brand evaluations predominantly concentrate on diagnosing the issue instead of proposing an effective remedy to counteract the bias stemming from recording devices (Adams et al., 2012; Rempel et al., 2013; P´erez-Granados et al., 2019).”

“Subsequently, traditional peak detection algorithms (such as proposed by Dumpala et al. (1982) or Sueur (2018)) are used to find prominent peaks.”

“Despite the importance of this issue, many works focus on the behavioral pattern of the index without considering its magnitude (Tuneu-Corral et al., 2020), losing reproducibility among studies.”

“Due to their flexibility, EAI have been used for different aims, such as acoustic niches, uniformity, soundscape degradation, and environmental studies (Towsey et al., 2014; Shaw et al., 2021; Benocci et al., 2020).”

“Due to their flexibility, EAI have been used for different aims, such as acoustic niches, uniformity, soundscape degradation, and environmental studies (Towsey et al., 2014; Shaw et al., 2021; Benocci et al., 2020).”

“The literature review (see Appendix A) highlights ecoacoustic data variability reduction strategies, yet it predominantly focuses on intra-brand evaluations using filtering methods and sampling adjustments for EAI Hyland et al. (2023a).”

“Ecological acoustic indices (EAI) are one of the most common metrics to study soundscapes as they are easy to calculate and are assumed to have biological interpretability (Buxton et al., 2018; Sueur and Farina, 2015; Farina and Gage, 2017).”

“Furthermore, RMS normalization does not limit the audio to a specific range, thus avoiding using other signal normalization techniques that could suppress relevant information (Vu et al., 2018).”

“The intra-study variability of EAI has been attributed to various factors, such as the sensitivity to changes in environmental conditions, noise, and the sampling protocol Bradfer-Lawrence et al. (2019a); Metcalf et al. (2021a); Lillis and Mooney (2022); Zwerts et al. (2022).”

“The intra-study variability of EAI has been attributed to various factors, such as the sensitivity to changes in environmental conditions, noise, and the sampling protocol Bradfer-Lawrence et al. (2019a); Metcalf et al. (2021a); Lillis and Mooney (2022); Zwerts et al. (2022).”

“To solve this issue, we employed the Kaiser resampling method proposed in (Liu et al., 2020) to equalize the sampling rates of the devices and we found that effectively decreased the recorder bias for ACIft and M.”