Bird population declines and species turnover are changing the acoustic properties of spring soundscapes

The study addresses how widespread declines in bird populations and species turnover affect the acoustic properties of natural soundscapes, a key pathway for human engagement with nature. In the context of urbanization and an 'extinction of experience,' reduced direct contact with nature, combined with global biodiversity loss, may be diluting the quality of interactions with nature. Sound is central to perceptions of environmental quality, and birds are major contributors to natural soundscapes. Reductions in avian abundance and species richness, alongside biotic homogenization, are likely to reduce acoustic diversity and intensity, potentially lowering the quality of nature experiences. However, relationships between community change and soundscape properties are nuanced and non-linear, as different species contribute differently to acoustic structure. The study’s purpose is to reconstruct historical soundscapes across North America and Europe to quantify long-term changes in acoustic diversity, evenness, intensity, and heterogeneity, linking them to changes in bird community richness and abundance.

Prior work links decreasing human-nature interactions with negative health and well-being outcomes and highlights the importance of natural sounds, especially birdsong, for nature connectedness and perceived environmental quality. Acoustic indices have been used to characterize diel and habitat-related variation in soundscapes and are broadly correlated with avian species richness and abundance. However, empirical evidence for long-term change in soundscape characteristics is limited due to a lack of historical recordings. Concurrently, literature documents declines in avian abundance and richness, increased biotic homogenization, and pervasive anthropogenic noise, all likely to degrade natural soundscapes. These insights motivate reconstructing soundscapes from monitoring data and archived recordings to overcome shifting baselines and to evaluate multi-decadal change.

Data sources: Annual bird count data were compiled from the North American Breeding Bird Survey (NA-BBS; 1996–2017) and the Pan-European Common Bird Monitoring Scheme (PECBMS; 1998–2018). NA-BBS routes comprise 50 sites with 3-minute point counts, while PECBMS aggregates multiple national schemes using line transects, point counts, or territory mapping. Records of subspecies were aggregated to species level; hybrids and genus-only records were removed. Only sites surveyed at least three times in the study period were retained.

Sound recordings: For all species detected, high-quality (category A) recordings categorized as song, call, or drumming were downloaded from Xeno-canto, restricted to recordings from the corresponding continent. Files >30 s were preferred; if unavailable, shorter files meeting criteria were stitched to exceed 30 s. If >50 files met criteria for a species, 50 were randomly selected. Each file was standardized: clipped to 25 s (from 2.5–27.5 s of the original), resampled at 44.1 kHz, normalized (+6 dB), mono mp3. Species lacking any qualifying recordings led to exclusion of the sites where they were detected (<1.5% NA-BBS sites; <3.5% PECBMS sites).

Soundscape reconstruction: For each site-year count file, a 5-minute soundscape was constructed. For each individual recorded, a randomly selected 25 s species recording was inserted at a random time with playback volume randomly drawn from a uniform distribution to represent variable proximity. This was repeated for all individuals of all species. Each site-year soundscape construction was iterated five times, and acoustic indices were averaged across iterations. One mp3 per site (fifth iteration) was saved for PECBMS sites and the first NA-BBS site per route.

Acoustic indices: Four indices quantified soundscape characteristics: Acoustic Diversity Index (ADI; Shannon diversity across frequency bands), Acoustic Evenness Index (AEI; Gini-based evenness across frequency bands), Bioacoustic Index (BI; area under dB spectrum up to 22,050 Hz), and Acoustic Entropy (H; product of spectral and temporal entropy). Defaults were used except BI maximum frequency set to 22,050 Hz. Lower ADI, BI, and H and higher AEI indicate reduced acoustic diversity/intensity.

Simulations and validation: Constructed soundscapes were generated for simulated communities to verify index behavior. (1) Single-species communities with 1–10 individuals; (2) multi-species communities with 2, 3, 4, 5, 10, 20, or 50 species and 1–10 individuals per species (70 combinations), iterated 100 times per combination for each continent’s species pool. GLMs related indices to log(number of individuals), log(number of species), and their interaction.

Statistical analysis of trends: For observed data, annual site-level values of each index were standardized within site (z-scores) to control for detectability, observer effects, and initial community structure. Gaussian GLMMs (lme4) were fitted separately for North America and Europe with standardized index as response and fixed effects of latitude, longitude, and year (continuous). Random effects: site and year; plus state and route (North America) or country (Europe). Likelihood ratio tests assessed fixed effects. Models were also fitted with year as categorical for annual predictions. Geographic variation in temporal trends was explored with latitudeyear and longitudeyear interactions.

Community trends and their association with indices: GLMMs equivalent in structure modeled standardized annual total number of individuals and species against latitude, longitude, and year. Site-level linear trends in individuals, species, and each acoustic index were estimated via GLMs with year as predictor. Separate GLMMs related site-level trends in each index to site-level trends in individuals and species (and their interaction), with random intercepts for state (North America) or country (Europe). Significance robustness was evaluated via 1000 bootstrap datasets sampling trend estimates by their standard errors.

Sensitivity analyses: For 1000 simulated 10-species communities declining from 10 to 5 individuals per species over 6 years, four reconstruction variants were compared: standard 5-minute with uniform volume; 3-minute (greater overlap); 10-minute (less overlap); 5-minute with half-normal volume distribution (more distant vocalizations). Methodological choices affected absolute index values but not the relative temporal responses to community change.

Software: Sound processing via SoX; indices via R packages seewave, soundecology, tuneR in R v3.5.1. Spatial autocorrelation (Moran’s I) was negligible.

• Across >200,000 sites, indices indicate chronic deterioration in soundscape quality since the late 1990s. North America and Europe show significant declines in ADI, BI, and H, with a significant increase in AEI in North America, implying more homogeneous and quieter soundscapes.
• GLMM fixed effects for Year: North America: ADI estimate −0.00198 (p=0.004); AEI +0.00144 (p=0.011); BI −0.00217 (p=0.014); H −0.00625 (p<0.001). Europe: ADI −0.0073 (p=0.014); AEI +0.0053 (p=0.079, not significant); BI −0.0056 (p=0.009); H −0.0095 (p=0.002).
• Geographic patterns: Greatest reductions in acoustic diversity (decreased ADI, increased AEI) in northern and western areas of both continents; BI declines strongest in northern and eastern North America; European BI shows no clear spatial pattern; H decreased more in eastern North America and slightly more in the south; in Europe, H decreased in north and west, with slight increases toward south and east.
• Community structure is declining: Significant declines in total individuals (NA-BBS and PECBMS) and in total species (NA-BBS) over 25 years.
• Strong associations between site-level trends in community metrics and acoustic indices (Table 4): Positive relationships between trends in species richness and total individuals with ADI, BI, and H; negative with AEI. For North America, effects of species on ADI (estimate 0.578), BI (0.615), H (0.662) and of individuals on ADI (0.109), BI (0.421), H (0.910) were all highly significant (p<0.001). Europe showed similarly strong positive effects (e.g., species on H 1.095; species on BI 0.947; p<0.001). Interactions were generally small or mixed.
• Simulations confirmed expected index behavior: Increasing abundance and richness increases ADI, BI, H and decreases AEI, with diminishing sensitivity at high richness/abundance.
• Despite overarching declines, substantial site-level variation exists, with all combinations of index increases/decreases observed; correlations among indices are generally positive among ADI, BI, H and negative between AEI and the others.

The findings demonstrate that widespread declines in bird abundance and richness translate into measurably poorer spring soundscapes—less diverse, more even (dominated by fewer frequency bands), and quieter—across North America and Europe. This addresses the research question by empirically linking community change to acoustic properties using reconstructed historical soundscapes at continental scale. The significance lies in revealing a chronic deterioration in a key sensory pathway through which people engage with nature, with implications for mental health and well-being and potential feedbacks to conservation engagement. Geographic heterogeneity in trends likely reflects interacting drivers at regional to local scales—changes in richness and abundance, taxonomic/functional/phylogenetic diversity, and compositional turnover—combined with differential environmental pressures. While indices correlate strongly with community trends, local context (initial community structure and species-specific vocal traits) modulates acoustic responses, making soundscape dynamics multidimensional. Anthropogenic noise and broader biodiversity declines in other taxa likely exacerbate the deterioration, further reducing acoustic bandwidth and masking natural sounds. Methodological sensitivity analyses and site-level standardization suggest the temporal trends reported are robust to reconstruction choices, though absolute index values depend on construction parameters. Overall, the results underscore the need to integrate soundscape quality into conservation policy and public health considerations.

This study introduces a scalable, data-driven approach to reconstruct historical soundscapes by integrating large-scale bird monitoring with archived recordings, enabling quantification of multi-decadal changes in acoustic diversity, evenness, intensity, and entropy. It documents pervasive declines in soundscape quality across North America and Europe linked to decreasing avian richness and abundance, with important implications for human-nature connections and well-being. Future directions include: (1) expanding systematic collection and curation of field soundscape recordings (e.g., via autonomous recorders and citizen science) to capture anthropogenic noise and multi-taxon contributions; (2) incorporating species-specific vocalization frequency and duration into reconstructions; (3) forecasting future soundscapes under environmental change scenarios; (4) mapping soundscape quality changes to human population distributions and access points to assess exposure and health impacts; and (5) ensuring conservation policy protects and restores high-quality natural soundscapes to avoid negative feedback loops between diminished experiences and reduced conservation support.

• Reconstructed, not recorded, soundscapes: While avoiding shifting baselines, reconstructions rely on assumptions (e.g., one 25 s clip per individual, random insertion/volume) and lack species-specific vocalization rate and duration data, potentially affecting absolute index values.
• Background noise in source recordings: Despite using high-quality files and clipping, some residual ambient noise may influence indices, though expected to be random and to reduce, not inflate, detection of trends.
• Methodological parameter choices (soundscape length, volume distribution) influence absolute metrics; sensitivity analyses indicate relative temporal trends are robust.
• Variability in survey methods across PECBMS schemes; site-level standardization aims to control for this.
• Spatial autocorrelation in residuals was statistically significant but negligible in magnitude; not modeled explicitly beyond random effects.
• Not all species had qualifying recordings, leading to exclusion of a small subset of sites.
• Indices respond nonlinearly and are context-dependent; site-level variation means some local trends deviate from continental averages.