The SARS-CoV-2 pandemic drastically changed daily life. Lockdown measures resulted in reduced traffic mobility and, subsequently, a changed acoustic environment. The exceptional lockdown was used to analyze its impact on the urban acoustic environment using ecoacoustic indices. Using data from 22 automated sound recording devices located in 9 land use categories (LUCs) in Bochum, Germany, the normalized difference soundscape index (NDSI) and Bioacoustics index (BIO) were explored. The NDSI quantifies the proportion of anthropophonic to biophonic sounds, and BIO quantifies the total sound activities of biological sources. The mean differences and standard deviation (SD) were calculated 5 weeks before and 5 weeks during the first lockdown. Pronounced peaks for the NDSI and BIO before lockdown that diminished markedly during lockdown were observed, however, with distinct differences in terms of the LUC. The mean NDSI increased from 0.00 (SD = 0.43) to 0.15 (SD = 0.50), the mean BIO decreased from 4.74 (SD = 2.64) to 4.03 (SD = 2.66). Using the NDSI and BIO together reveals that changes of the acoustic environment during lockdown are mainly driven by decreased anthropophonic sound sources. These results suggest that further studies are needed to tailor ecoacoustic indices more accurately to conditions of the urban environment.
I. INTRODUCTION
In the end of 2019, a cluster of patients with pneumonia was observed in Wuhan, China, which turned out to be caused by a new coronavirus type, SARS-CoV-2 (Chan et al., 2020). The disease, named COVID-19 by the World Health Organization, spread rapidly since then with more than 30 × 106 cases in more than 180 countries (Guarner, 2020) and evolved to a global pandemic in early 2020 (Cucinotta and Vanelli, 2020). The disease has become a threat to life and health for many people (Guarner, 2020; Ornell et al., 2020). Thus, to mitigate the rapid spread of COVID-19, there have been measures taken in many countries that limit direct contact of people to decrease the chance of transmission. In Germany, strict lockdown measures resolved around mid-March 2020 included the closure of schools, restaurants, shops, leisure facilities, and sports clubs, as well as restrictions on visits to hospitals and nursing care facilities (Ministerium für Arbeit GuSN-W, 2020). Additionally, social contacts were restricted, allowing only a limited number of private meetings. Consequently, the mobility of an entire society has been drastically reduced, which is reflected in the traffic counting data published by the ministry for traffic in North Rhine-Westphalia (NRW), as traffic volumes reduced by around 40% in April 2020 compared to those in April 2019 (Straßen.NRW, 2020).
Despite the dire consequences of the pandemic, however, the profound measures allowed for natural experiments on an enormous spatial and temporal scale that would otherwise not have been feasible. For example, various research groups have studied the changes in noise levels in European cities during lockdown periods. Interestingly, they all observed decreases in decibel levels ranging from 3 to 10 dB(A) (Aletta et al., 2020; Basu et al., 2021; Hornberg et al., 2021; Rumpler et al., 2020; Smith et al., 2020).
In addition to examining the impact of mobility constraints on noise, the scenario provides a unique opportunity to examine other qualities and aspects of the urban acoustic environment beyond noise. Soundscape research investigates precisely these qualities, which go far beyond noise. This research field includes various concentrations, such as cognitive-psychological aspects of acoustic environments, as well as primarily physical-acoustic characteristics from biological, geophysical, and anthropogenic sounds emanating from landscapes, known as soundscape ecology (Aletta et al., 2018; Brooks et al., 2014; Moebus et al., 2020; Pijanowski et al., 2011; Van Kamp et al., 2016; Yang and Kang, 2005).
Changes in the perception of soundscapes during lockdowns have been studied. For example, an online survey from Lyon, France, revealed that not only perceived noise has decreased, but natural sounds were also perceived better (Acoucité, 2020). Sound pressure-based analyses in San Francisco, CA, focussing mainly on background and ambient noise, have shown that reduced traffic volumes have caused bird songs to change and become more dominant in the acoustic environment (Derryberry et al., 2020).
In a study from Spain, Lenzi et al. (2021) applied ecoacoustic indices, such as the normalized difference soundscape index (NDSI), acoustic complexity (ACI), and acoustic richness (AR), in one urban site during lockdown and release of restrictions lasting from March 15 to June 21. The results showed that perceived eventfulness, ACI, and AR increased over the study period while the number of perceived technological sounds decreased. Since this study was performed at only one single site, further studies determining the impact of human activities on urban acoustic environments with respect to different land use types (LUTs) are needed.
Accordingly, the aim of our study was to investigate the impact of the lockdown measures on patterns of biophonic and anthropophonic sounds in nine different types of urban fabric, taking the opportunity to use the exceptional lockdown scenario with its great changes in traffic volumes. Using longitudinal automated audio recordings we applied metrics from soundscape ecology in the city of Bochum, Germany. Bochum is located in the Ruhr metropolitan area, which has a population of around 5.1 × 106, making it the largest urban agglomeration in Germany and the fifth largest in Europe. It represents a highly urbanized area, whose acoustic environment is strongly influenced by vehicular traffic.
For our purposes, we explored the NDSI and the bioacoustic index (BIO), as they are intended to represent bio- and anthropophonic activity. Thus, we can observe characteristics of ecoacoustic indices in different land uses and their changes during the lockdown with its drastic decreases in human activity. The NDSI quantifies the proportion of sounds caused by humans and biological sources, whereas the BIO serves as a measure of total biophonic activity. In doing so, we pursue two aims: First, we investigate the relationship between reduced human activity and the development of NDSI and BIO. Second, we apply these indices, which are otherwise primarily applied in the field of soundscape ecology, to the urban environment in different locations over an expanded observation period. This should provide further insight into the application of ecoacoustic indices in the study of the urban acoustic environment.
II. METHODS
A. Sound data
For our analyses, we used the comprehensive data set of the SALVE (acoustic quality and health in urban environments) project, described in detail in Haselhoff et al. (2022). Briefly, the aim of SALVE is to measure spatial-temporal differences in acoustic environments using manual and automatic devices at more than 700 different urban locations in the city of Bochum, Germany. The ongoing study started in May 2019 and, thus, includes recordings before and during the first COVID-19 lockdown period in Germany. Audio recordings were performed using 24 Wildlife Acoustics SM4 acoustic recorder devices (Maynard, MA). Due to technical difficulties during the lockdown period, two devices were excluded. Overall, we sampled mono-recordings, saved as .wav (waveform audio) files, at 44 100 Hz with a bit-depth of 16. The devices recorded 3-min samples every 26 min, corresponding to 50 recordings per day.
Recording sites were sampled based on different LUTs to reflect a wide range of acoustic environments that are typical for the urban environment. The distribution of our audio devices in the city of Bochum is mapped in the supplementary material.1 The initial sampling was based on the land use mapping provided by the Regional Association for the Ruhr Area (RVR; Regionalverband Ruhr, 2020). For our analyses, we defined land use categories (LUCs) on the base of LUTs of the RVR, characteristics of surrounding areas, and pictures of the surroundings of the recording devices (Fig. 1). The definition was carried out from each team member in case of disagreement, and a final category was assigned in a panel round.
(Color online) The LUCs and number (n) of devices per category. Example images for each category, taken at the location of the respective device, are shown. The satellite imagery for each LUC pictured is provided in the supplementary material.1
(Color online) The LUCs and number (n) of devices per category. Example images for each category, taken at the location of the respective device, are shown. The satellite imagery for each LUC pictured is provided in the supplementary material.1
B. Definition of the pre- and during lockdown phases
According to lockdown measures, we divided our observation periods into an interval before and an interval during the lockdown. We chose March 16 as the cutoff date for our analyses, the day on which the most important lockdown measures came into force. The Ministry of Labor, Health, and Social Affairs of North Rhine-Westphalia imposed entry bans on travelers returning from risk areas, community facilities, health care facilities, and schools for a period of 14 days after their stay and bans or restrictive limitations on visits to health care facilities beginning on March 16. In addition, on March 17, the closure of bars, clubs, theaters, cinemas and museums, gyms, swimming pools, saunas, adult education centers, music schools, sports clubs, other sports and recreational facilities, arcades, gambling halls, prostitution establishments, and the ban on all public events were implemented (Ministerium für Arbeit GuSN-W, 2020). On March 23, additional regulations went into effect that restricted meetings and gatherings of more than two people in public, among other things, with exceptions for relatives, spouses or domestic partners, household members, minors, mandatory meetings for business, professional, examination, or support reasons. In addition to the legal measures, the Robert Koch Institute (RKI), as the main governmental institution for public health protection in Germany, recommended to reduce social contacts and work from home if possible (Robert Koch-Institut, 2020). However, due to the non-legally binding nature of the RKI work at home recommendations, travel to work continued. A timeline of events can be found in the supplementary material.1 In summary, we set the lockdown period to 5 weeks, starting with the first day of lockdown measures in Germany; thus, the last measurement day was April 19. We defined the pre-lockdown period as 5 weeks before the start of the lockdown, which corresponds to the period from February 10 to March 15, 2020. The number of recordings by LUC, according to the lockdown periods, is presented in Table I.
The number of recordings by LUC, stratified by pre- and during lockdown.
LUC . | Pre-lockdown (n) . | During lockdown (n) . |
---|---|---|
Commercial area | 3500 | 3493 |
Green space | 3500 | 3492 |
Main street | 5250 | 5238 |
Play or sportsground | 1750 | 1746 |
Residential areas | 3500 | 3492 |
Residential street | 5250 | 5238 |
Small garden near house | 8750 | 8729 |
Urban agricultural land | 3500 | 3492 |
Urban forest | 3500 | 3492 |
Total | 38 500 | 38 412 |
LUC . | Pre-lockdown (n) . | During lockdown (n) . |
---|---|---|
Commercial area | 3500 | 3493 |
Green space | 3500 | 3492 |
Main street | 5250 | 5238 |
Play or sportsground | 1750 | 1746 |
Residential areas | 3500 | 3492 |
Residential street | 5250 | 5238 |
Small garden near house | 8750 | 8729 |
Urban agricultural land | 3500 | 3492 |
Urban forest | 3500 | 3492 |
Total | 38 500 | 38 412 |
C. Calculation of acoustic indices
For our analysis, we calculated the NDSI and BIO. The NDSI was introduced by Kasten et al. (2012) to measure the proportion of anthropogenic disturbance. It is a measure of the ratio of anthropophonic and biophonic activity. To calculate it, first, a power spectral density (PSD) of a signal is computed (Welch, 1967). This gives an estimate for the proportion of energy of a sound contained in specific frequencies. Here, 1 kHz wide frequency bins are used. Anthropophonic activity is assumed to have the most energy in the frequency range 1–2 kHz, and biophonic sounds are assumed to mostly occupy the frequency spectrum from 2 to 8 kHz. The NDSI is calculated as NDSI = (β − α)/(β + α), where β and α are the total estimated PSD in the biophonic and anthropophonic range, respectively (Kasten et al., 2012; Sueur, 2018). Therefore, α is a measure of the energy contained in the frequency range 1–2 kHz, and β gives the energy contained in the frequencies from 2 to 8 kHz. The NDSI values range from −1 to +1, as it is normalized due to the division by the sum of the powers in both sound sources. Here, a NDSI of −1 represents a recording with energy only in the frequencies 1–2 kHz, thus, assuming to contain only anthrophonic sound. For a NDSI of +1, the opposite is true, and the sound is assumed to be exclusively biophonic.
Boelman et al. (2007) established the BIO to assess avian abundance in ecosystems. It is a measure of the total biophonic activity. Here, it is also assumed that biophonic sounds mostly take up frequencies from 2 to 8 kHz. For this range, a frequency spectrum is calculated. The BIO is then computed by calculating the area under the curve for the spectrum (Boelman et al., 2007). For our recordings, we calculated both indices in R (Ihaka and Gentleman, 1996) using the soundecology package (Sueur, 2018; Villanueva-Rivera et al., 2018).
So far, the indices described above have mostly been applied in natural, nonurban areas. Therefore, their application in urban areas is not yet fully evaluated. Both indices assume that certain sound sources occupy specific frequency ranges. These assumptions do not always hold true in an urban environment. Fairbrass et al. (2017) found that various anthropophonic sounds are also present in frequencies over 2 kHz. Therefore, the authors recommend to pre-process sounds before calculating ecoacoustic indices. They conclude that “the NDSI could be used to measure the ratio of biophonic to anthrophonic activity” (Fairbrass et al., 2017). Devos (2016) concluded that these indices “show promising characteristics for use in urban soundscape characterization.” Asensio et al. (2020) recommend reporting these indices in a taxonomy proposal for studies about COVID-19-related changes of the acoustic environment. Thus, we used these indices despite their missing evaluation in the urban environment because application of ecoacoustic indices in urban areas still needs further evaluations.
D. Analysis methods
Ahead of our analyses, we checked our data for plausibility. We investigated the NDSI and BIO for outliers, defined as values > Q3 + 1.5 ∗ IQR, where Q3 represents the third quartile and IQR is the interquartile range. Then, we sampled randomly from the outlier recordings (n = 70) and checked manually for methodical failures and biasing sounds that clearly distort the sound analyses, but we did not find any flawed recordings.
The means and standard deviations (SDs) for the NDSI and BIO were calculated for the periods before and during the restrictions. Differences between the pre- and during lockdown phases were calculated as Δ = - , where represents the mean of during lockdown phase and is the mean of the pre-lockdown phase. We calculated the mean NDSI and BIO of all LUCs by day (referred to as NDSI24h and BIO24h.) Using the NDSI24h and BIO24h, we described the mean course of the indices for each day over the observation period. By calculating the mean NDSI and BIO of all LUCs for each of the 5-week periods pre- and during lockdown (referred to as NDSI35d and BIO35d), we conducted a comparison of the pre- and during lockdown periods. As our aim was a descriptive analysis, no t-tests or other significance tests were performed. Furthermore, as the data set at hand is very large, p-values go toward zero for sample sizes going toward infinity (Callegaro et al., 2019) and use of pass-fail significance reporting can confuse the interpretation of the findings (Reese, 2004). Line plots were created to illustrate the development of the indices over time. For these graphs, the indices were averaged by date to prevent overplotting. The line plots show the means of the indices and 95% confidence intervals for all devices and each LUC.
All calculations and plots were created using the software R, version 4.1.1 (Ihaka and Gentleman, 1996), the ggplot2 package (Wickham, 2016), and the ggpubr package (Kassambara and Kassambara, 2020).
III. RESULTS
A. NDSI and BIO for all LUCs
Overall, we found a distinct increase in the NDSI24h during our observation period. Starting at around −0.1, from February 2, the NDSI24h increases heavily with peaks, in particular, increasing noticeably on 27 February and 5 and 11 March (Fig. 2). The decline of the extent of negative peaks after March 16 leads to a mean increase of about 0.1 until the start of April and indicates a shift to more biophonic sounds. Around April 3, the NDSI24h increases again and is around 0.2. Alongside marked fluctuations with pronounced peaks, a slight increasing trend of the NDSI24h is noticeable overall. This impression is confirmed by an overall increase of +0.15, comparing NDSI35d before and after March 16 (see Table II). The trend seems to start before March 16.
The development of NDSI and BIO before and during lockdown phase is averaged per date over all LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
The development of NDSI and BIO before and during lockdown phase is averaged per date over all LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
The mean, standard deviation (SD; in brackets), and difference between during and pre-lockdown means for NDSI35d and BIO35d for all LUCs.
. | NDSI35d . | BIO35d . | ||||
---|---|---|---|---|---|---|
Index . | Pre () . | During () . | Δ . | Pre () . | During () . | Δ . |
LUC/lockdown phase . | Mean (SD) . | Mean (SD) . | - . | Mean (SD) . | Mean (SD) . | - . |
Commercial area | −0.41 (0.27) | −0.27 (0.33) | 0.14 | 5.81 (2.94) | 3.92 (1.90) | −1.89 |
Green space | −0.01 (0.44) | 0.31 (0.46) | 0.32 | 4.11 (2.31) | 3.90 (2.78) | −0.21 |
Main street | −0.18 (0.38) | −0.37 (0.31) | −0.19 | 6.18 (2.44) | 5.11 (1.80) | −1.07 |
Play or sportsground | −0.11 (0.32) | 0.10 (0.40) | 0.21 | 4.64 (2.61) | 3.21 (2.43) | −1.43 |
Residential areas | 0.14 (0.45) | 0.39 (0.43) | 0.25 | 4.55 (2.66) | 3.82 (2.80) | −0.73 |
Residential street | −0.09 (0.38) | −0.04 (0.44) | 0.05 | 4.68 (2.23) | 3.79 (2.12) | −0.89 |
Small garden near house | 0.15 (0.42) | 0.34 (0.45) | 0.19 | 4.79 (2.52) | 4.50 (3.13) | −0.29 |
Urban agricultural land | 0.12 (0.31) | 0.32 (0.36) | 0.20 | 3.87 (2.38) | 3.27 (2.50) | −0.60 |
Urban forest | 0.28 (0.36) | 0.63 (0.28) | 0.35 | 3.24 (2.52) | 3.22 (3.01) | −0.02 |
Total | 0.00 (0.43) | 0.15 (0.50) | 0.15 | 4.74 (2.64) | 4.03 (2.66) | −0.71 |
. | NDSI35d . | BIO35d . | ||||
---|---|---|---|---|---|---|
Index . | Pre () . | During () . | Δ . | Pre () . | During () . | Δ . |
LUC/lockdown phase . | Mean (SD) . | Mean (SD) . | - . | Mean (SD) . | Mean (SD) . | - . |
Commercial area | −0.41 (0.27) | −0.27 (0.33) | 0.14 | 5.81 (2.94) | 3.92 (1.90) | −1.89 |
Green space | −0.01 (0.44) | 0.31 (0.46) | 0.32 | 4.11 (2.31) | 3.90 (2.78) | −0.21 |
Main street | −0.18 (0.38) | −0.37 (0.31) | −0.19 | 6.18 (2.44) | 5.11 (1.80) | −1.07 |
Play or sportsground | −0.11 (0.32) | 0.10 (0.40) | 0.21 | 4.64 (2.61) | 3.21 (2.43) | −1.43 |
Residential areas | 0.14 (0.45) | 0.39 (0.43) | 0.25 | 4.55 (2.66) | 3.82 (2.80) | −0.73 |
Residential street | −0.09 (0.38) | −0.04 (0.44) | 0.05 | 4.68 (2.23) | 3.79 (2.12) | −0.89 |
Small garden near house | 0.15 (0.42) | 0.34 (0.45) | 0.19 | 4.79 (2.52) | 4.50 (3.13) | −0.29 |
Urban agricultural land | 0.12 (0.31) | 0.32 (0.36) | 0.20 | 3.87 (2.38) | 3.27 (2.50) | −0.60 |
Urban forest | 0.28 (0.36) | 0.63 (0.28) | 0.35 | 3.24 (2.52) | 3.22 (3.01) | −0.02 |
Total | 0.00 (0.43) | 0.15 (0.50) | 0.15 | 4.74 (2.64) | 4.03 (2.66) | −0.71 |
Regarding the BIO24h, Fig. 2 reveals a sharp overall decrease during the lockdown. In Table II, we can see that, in fact, the BIO35d decreased, on average, by −0.71. Interestingly, contrary to the NDSI35d, the BIO35d reduction suggests that biophonic sounds decreased during lockdown. Similar to the NDSI24h, we observed great variations of the BIO24h pre-lockdown, which during the lockdown, turned out to be considerably lower. The decrease occurs a few days before March 16 and decreases even more just before March 22. Following, we can observe an interesting weekly pattern during lockdown that was not pronounced pre-lockdown: On the Sundays on March 29, April 5, and April 19, the BIO24h is noticeably lower compared to the days around them. An exception is observed on the Easter holiday weekend, on which the BIO24h decreases sharply on April 13 (Monday), which is also a legal holiday in Germany, but not on the Sunday before.
B. NDSI and BIO stratified by LUCs
The overall increase in NDSI24h observed during the lockdown is also present in most single LUCs (Fig. 3). However, in the LUC “main street,” the NDSI24h drops relatively sharply just before the implementation of the restrictions after a slight peak and then remains at a relatively low level. Furthermore, Table II shows that in main street, the NDSI35d decreased from the second lowest measured value of all LUCs pre-lockdown (−0.18) to the lowest measured value during lockdown (−0.37). This result fits well to the pattern across LUCs in Fig. 3, where only in the LUC main street does the NDSI24h decrease.
The development of NDSI before and during lockdown phase is averaged per date and stratified by the LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
The development of NDSI before and during lockdown phase is averaged per date and stratified by the LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
In contrast to the NDSI24h, the BIO24h in main street is similar to the overall course of BIO24h across all LUCs especially with respect to the peaks pre-lockdown. We also observe a generally reduced fluctuation of the BIO24h during lockdown as compared to before lockdown (Fig. 4). Compared to other urban areas, commercial areas probably imply lower levels of biophonic activity as these locations are rather busy with human activity, traffic, and machine noise. However, we measured the second highest BIO35d (5.81) and highest reductions (−1.89) in the “commercial area” (Table II). Furthermore, the commercial area NDSI35d had the lowest value pre-lockdown (−0.41), which increased moderately during the lockdown (+0.14). An additional striking pattern appears in the commercial area and “residential area” by showing low peaks of NDSI24h on March 22, which subsequently drop and rise again just before April 6 (Fig. 3). Interestingly, April 6 was the start of the school holidays in Bochum, which likely further reduced traffic volume. This could explain the shift toward biophonies as indicated by the NDSI24h.
The development of BIO before and during lockdown phase is averaged per date and stratified by LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
The development of BIO before and during lockdown phase is averaged per date and stratified by LUCs. The vertical lines represent the 95% confidence interval. The solid vertical line represents the start of first restrictions on 16 March and the dashed vertical line represents the start of strict contact restrictions starting on 22 March.
The development of the BIO24h is similar in most of the LUCs, except for “green space” (Fig. 4). In this LUC, the BIO24h shows minimal changes pre- and during the lockdown, as well as no pronounced peaks between February 24 and March 10.
IV. DISCUSSION
The aim of this paper is to present and discuss results of an application of ecoacoustic indices NDSI and BIO in urban spaces to evaluate the impact of the lockdown episodes on the urban acoustic environment. The profound preventive measures taken due to SARS-CoV-2 allow for a natural experiment that simply would not have been feasible otherwise. In particular, the effects of reduced human activities, which are mostly due to road and air traffic, offer the opportunity for a detailed study of their impact on the acoustic environment. Our results reveal specific patterns in the ecoacoustic indices used with respect to different LUCs and in terms of different activities experienced during the pre- and lockdown phases. The overall increased NDSI35d during the lockdown period suggests a shift in the proportion of anthropophonic and biophonic sounds toward biophonies. However, the increase between the pre- and during lockdown periods varied between LUCs. For instance, the highest increases of the NDSI35d occurred in “urban forest,” green space, and residential areas, whereas in main street, a decrease of −0.19 was observed. According to the BIO35d, we observed an average decrease of about −0.71, suggesting that biophonic activity has decreased at all measurement locations during lockdown compared to during the pre-lockdown period. The highest reduction was found in LUCs strongly influenced by human activities, such as the commercial area, “play- or sportsground” and main street. These observations are certainly worthy of further intensified analyses.
We previously analysed noise levels during the lockdown in the city of Bochum, revealing that the noise reductions on main streets were present, albeit less, in comparison to green spaces and urban forests (Hornberg et al., 2021). This suggests that despite the reduced activities resulting from the lockdown measures, the traffic volume on main streets remained substantial. Respectively, as the NDSI is a measure to estimate the level of anthropogenic impact on the soundscape (Kasten et al., 2012), this might explain our observed decrease in NDSI35d in the LUC main street. This result emphasizes the relevance of considering the type of land use in the selection of ecoacoustic indices for the analysis of urban acoustic environments. The overall increase in the NDSI35d is in line with the findings from Derryberry et al. (2020) and Acoucité (2020). This was expected, considering the fact that traffic volume and human activity was drastically reduced during the lockdown period. An impact of the LUT on the acoustic environment is often assumed, but quantitative evidence is still scarce. Recently, Dein and Rüdisser (2020) reported that the proportion of biophony can be predicted by landscape characteristics like land cover as well as the “distance to nature” (D2N) index. Our results provide further empirical evidence as we also observed that changes of the acoustic indices pre- and during lockdown measures differed according to the LUC. For instance, a decrease in anthropophonic activity was especially clear for rather natural areas like urban forest and green space, whereas at main street, the rate of biophonic to anthropophonic sounds even shifted toward anthropophony.
One possible explanation for the rather unexpected result of decreased biophonic activity is reduced sound pressure levels of bird songs, which is supported by Derryberry et al. (2020), who indeed measured lower bird song amplitudes during the COVID-19 restrictions. However, another explanation can be derived from our findings: Anthropophonic sound sources also occupy frequencies above 2 kHz (Bradfer-Lawrence et al., 2019) and are, therefore, occupying the same frequency ranges as biophonic sources. Furthermore, they are most likely louder (higher amplitude) than biophonic sources. Accordingly, based on the observation of a generally considerable reduction in traffic volume during the lockdown, we could expect a large reduction of power at anthropophonic frequencies (1–2 kHz). However, at the same time, a reduction in higher frequencies above 2 kHz, albeit not as pronounced, can be expected. In terms of the NDSI, this means an increase in favour of biophonic components. Simultaneously, the BIO (which only considers frequencies >2 kHz) decreases. This is due to a considerable decrease in traffic as a prevalent anthropophonic sound source, which also occupies frequencies >2 kHz. Here, the BIO provides additional information about whether the NDSI has changed as a result of an increase or decrease in biophonic or anthropophonic sounds. Future research and urban sound analysis should consider applying these and additional (ecoacoustic) indices to gather information about the acoustic environment from multiple perspectives, as we have found useful in past studies to differentiate the nuanced differences between urban mixed use and urban forest (Lawrence et al., 2022). Our results emphasize the need for further evaluations of ecoacoustic indices in urban areas. A first step could be to adjust the lower frequency threshold for biophonic sounds to account for possible anthropophonic influences for the area of interest. In our experience, changing the threshold for biophonic activity from 2 to 4 kHz might be appropriate for most urban areas, but this assertion needs further investigation.
V. STRENGTHS AND LIMITATIONS
As in all research, this study has several strengths and limitations. One of the strengths proved to be the high spatial resolution measurements with 22 recording stations in different urban land uses. Unlike many other studies of the urban acoustic environment, our approach includes residential areas where people spend most of their time. The high temporal resolution, as well as the long-term measurements, are further strengths of this research. Also, noteworthy, is the systematic approach to collecting high temporal resolution, taking into account possible temporal variations. The implementation of a quality management procedure, which includes the application of a study protocol, qualification and training of field personnel, extensive plausibility checks of the assessed sound recordings, and calculated indices, increases the quality of the results.
Besides these strengths, limitations have to be pointed out as well. As mentioned previously, caution needs to be taken when interpreting ecoacoustic indices in urban areas as the classification into sources based on the frequencies of sounds is not always accurate (Fairbrass et al., 2017). Nevertheless, our analysis of outliers yielded results that lead to the assumption that these indices are, to some degree, usable in urban environments (see Sec. II D), as recordings with very high BIO and NDSI values do indeed mostly contain loud biophonic sounds. Attenborough et al. (2006) advise to consider the effect of weather conditions on noise measurements and ecoacoustic indices. For example, snow is known to alter sound propagation in a way that is perceived and measured to be quieter (Attenborough et al., 2006). Wind is also known to affect microphone measurements in rather low frequencies (van den Berg, 2006). Thus, it could affect not only noise measurements but also ecoacoustic indices. As small-scale weather data were not available, effects of wind and rain on the indices could not be taken into account. However, we can exclude possible influences of snow on the acoustic environment because it did not snow during the observation period. Another limitation is that our recordings started in May 2019 and we, therefore, had no control data for seasonal changes.
VI. CONCLUSION
The soundscape approach facilitates using information about urban sound beyond noise levels to try to shape acoustically pleasant spaces. After anecdotes of wildlife reclaiming the city and bird song becoming more prominent during COVID-19 lockdown measures, it could be demonstrated that the acoustic environment indeed changed. By using ecoacoustic indices, we could show that the progression of NDSI and BIO is different in the observed LUCs and the change can be attributed mainly to a decrease in anthropophonic sound sources rather than an increase in biophonic sounds. Only by considering the NDSI together with the BIO, were we able to draw the latter conclusion, thus, showing the importance of using multiple acoustic indices that complement each other when analysing the urban acoustic environment.
ACKNOWLEDGMENT
T.H. and J.H. contributed equally to this work.
See supplementary material at https://www.scitation.org/doi/suppl/10.1121/10.0013705 for figures of recording locations, satellite imagery, and the lockdown timeline.