Lowland fluvial oxbow lakes and their riparian zones are integral floodplain habitats providing key ecosystem services, including water quality maintenance, biodiversity support, and carbon sequestration (Ward 1998; Naiman et al. 2010; Graziano et al. 2022; Das et al. 2025). Yet across Europe and North America, 50–70% of floodplains have been altered by channelization, land use change, and agricultural pressures (Tockner and Stanford 2002; Tockner et al. 2022), and an estimated 90% of European riparian forests have been lost, with remaining fragments often in a critical state (Hughes et al. 2008). These changes have disrupted nutrient cycling and hydrological function (Carlson Mazur et al. 2022), reduced wildlife habitat and ecological corridors (Sweeney et al. 2004), and driven shifts in community composition across trophic levels (Paillex et al. 2007). Riparian restoration is therefore increasingly recognized as essential for mitigating these pressures (Matzek et al. 2015; Rodríguez-González et al. 2022) and for advancing policy frameworks such as the EU Water Framework Directive and UN Sustainable Development Goals 6 and 15. Robust long-term monitoring underpins adaptive management of these dynamic systems (Rodríguez-González et al. 2022).

Passive acoustic monitoring (PAM) has rapidly emerged as a non-invasive, scalable tool for biodiversity assessment, particularly for vocally active vertebrates (Sueur and Farina 2015; Gibb et al. 2019). PAM data can be analysed at the species level, manually, or using automated recognisers, or summarised across the entire soundscape using acoustic indices: numerical descriptors of biophony, anthrophony and geophony (Pijanowski et al. 2011; Sueur and Farina 2015). Acoustic indices have become a popular community-level approach because they reflect acoustic community patterns and scale to large datasets, but they show inconsistent performance as biodiversity proxies (Alcocer et al. 2022; Allen-Ankins et al. 2023; Bradfer-Lawrence et al. 2023, 2024; Metcalf et al. 2023), and their interpretation is sensitive to recording protocol, ecosystem and recorder hardware (Eldridge et al. 2018; Bradfer‐Lawrence et al. 2023). For this reason, recent guidance recommends complementing index analyses with at least partial species-level information, whether through manual aural inspection, expert spectrogram review, or automated recognisers, to ensure that index trends can be interpreted ecologically (Bradfer‐Lawrence et al. 2023, Hoefer et al. 2023). Long or continuous recording periods further improve PAM performance by capturing species and acoustic patterns that shorter, targeted windows miss (Hoefer et al. 2023). Within this framework, our study uses the established Acoustic Complexity Index (ACI) and the more recent Mid-Frequency Cover (MFC), the latter of which has been less extensively validated in temperate environments, anchored by manual aural inspection of a stratified subsample, applied year-round to a network of riparian oxbow lakes.

Despite their ecological importance, temperate riparian soundscapes, and oxbow lakes in particular, remain understudied with PAM. A recent systematic review confirmed that freshwater ecosystems are markedly underrepresented in soundscape research (Turlington et al. 2024). Among the few PAM studies conducted in oxbow systems specifically, recordings have been limited to the breeding season (Shaver et al. 2022). This breeding-season focus reflects a broader pattern in PAM research: comprehensive year-round sampling remains rare across terrestrial datasets globally (Darras et al. 2025), even though acoustic indices respond differently across temporal and frequency scales, and short or breeding-season-only sampling can mask ecologically meaningful patterns such as winter or off-season choruses (Metcalf et al. 2020; Bradfer-Lawrence et al. 2023). Capturing the full annual cycle is therefore essential not only for understanding biophonic phenology in understudied systems like oxbow lakes, but also for evaluating the temporal reliability of acoustic indices themselves. To our knowledge, this is the first year-round PAM assessment of temperate oxbow lakes; we conducted it in Central Europe, in the artificial oxbow systems of the Morava River floodplains — a heavily modified part of the Upper Danube region that nonetheless retains biologically rich stagnant-water habitats within a cultural landscape. Rather than relying on species identity or presence/absence data alone, we applied acoustic indices to capture diel and seasonal patterns in biophonic activity. While such indices are influenced by species richness and abundance, they also reflect ecological and behavioral drivers including resource availability, animal condition, and anthropogenic pressure (Bradfer‐Lawrence et al. 2024), we therefore treat them as integrative metrics of the acoustic environment, particularly of processes related to acoustic communication (Znidersic and Watson 2022).

Specifically, we examined: (1) the temporal patterns of the established Acoustic Complexity Index (ACI) and the Mid-Frequency Cover (MFC)—as proxies for biophonic activity and bird species richness across seasonal and diel cycles, and assessed how closely each index tracks aurally verified avian assemblages; and (2) their sensitivity to habitat condition and anthropogenic gradients, by testing whether protected and unprotected oxbow lakes differ in avian species richness, acoustic index values, and the temporal structure of biophonic activity. We hypothesised that (i) the two indices would differ in their capacity to track bird species richness, given their differences in frequency range and the acoustic properties they capture, and (ii) if conservation measures support more complete and undisturbed acoustic communities, protected sites would exhibit higher species richness and higher values of biophony-sensitive indices than otherwise comparable unprotected sites.

Study site

The study sites are situated in the lowlands of the Lower Morava Valley, in the eastern part of Czechia. The Morava River holds considerable significance in Europe for its historical role in colonization (continuously inhabited from sixth century AD), its impact on sediment transport (Kadlec et al. 2015), water management policies (particularly relevant in the context of climate change), and its rich biodiversity, especially within the Lower Morava River Basin (Maděra et al. 2013). In the first half of the twentieth century, the main channel of the river was straightened and shortened by approximately 40%, resulting in the creation of over 180 artificial oxbow lakes. These lakes serve as natural reservoirs for pisciculture and recreation. Some have been filled in and a few were repurposed for waste deposition. We selected seven man-made oxbow lakes situated on both sides of the twelve kilometre-long river segment near the city Uherské Hradiště, the centre of the cultural region of Moravian Slovakia (Fig. 1). The sites with an average altitude of 175 m are situated in intensively managed cultural landsape with remnants of fragile riparian forests. The distance between neighboring sites ranges from 0.9 to 2.7 km. Lakes exhibit varying characteristics in terms of site conservation, land use and land cover in its surrounding. Four of the study sites (Site 1 Tun, Site 4 Certak, Site 5 Konovy, and Site 6 Kanada) are protected simultaneously under the national Act no. 114/1992 Coll., on Nature and landscape protection and under the NATURA 2000 network (particularly Special Areas of Conservation - SAC) defined by European Commission Habitats Directive (92/43/EEC). Additionally, all study sites are located within a transregional biocorridor of the Territorial System of Ecological Stability (§ 4 (1) of Act No. 114/1992 Coll., on Nature and Landscape Protection). The conservation of these sites is closely tied to the protection of the freshwater fish Amur bitterling (Rhodeus sericeus amarus), several species of amphibians (e.g. Rana spp., Hyla arborea) and water insect (Odonata, Epitheca bimaculata in particular) and the floating aquatic plant water chestnut (Trapa natans). Three of these sites (Site 1, 5 and 6) are located within riparian forest or at its edge, featuring a riparian buffer zone with mature trees and dense undergrowth.

Environmental variables

To characterize the local environmental context surrounding each oxbow lake, we established a 300 m buffer around the open-water perimeter of each lake using QGIS 3.30.138, and quantified land-cover variables within this buffer from current orthophoto basemaps. Within each buffer we extracted the proportions of forest cover, arable land, and urban area. The selected buffer distance represents a compromise between acoustic, ecological, and landscape considerations. First, it approximates the effective acoustic sampling space of autonomous recording units. Although maximum detection distances vary widely among species and habitat types (Yip et al. 2017; Darras et al. 2018), detectability of most vocalizations in forested and semi-open habitats declines markedly beyond ~ 150 m, while larger forest birds are detectable up to 300–400 m (Winiarska et al. 2024). Because acoustic detection is directional and constrained by recorder position, the buffer should be interpreted as an approximation of the landscape influencing the recorded soundscape rather than a precise acoustic footprint. Second, this scale falls within the 125–500 m range shown to best predict floodplain bird community structure, with finer scales often providing higher explanatory power (Ónodi et al. 2024). Third, land cover at this scale captures broader environmental pressures affecting oxbow systems, including agricultural disturbance, edge effects, runoff, and noise propagation. The use of a single intermediate buffer was further justified by the small size and linear arrangement of the studied oxbows, which limited the applicability of nested buffers due to spatial overlap among sites.

Next, we included oxbow size, distance to the nearest road and road type (Class I & II roads - First Class Roads - intended mainly for long-distance and interstate transport, and Second Class Roads - intended for transport between districts) site conservation status, as a binary variable (Natura 2000 SAC vs. unprotected; based on national Act No. 114/1992 on the Conservation of Nature and Landscape and NATURA 2000 protection; see list of variables in Table 1; Fig. 2). Our seven sites with their adjacent riparian habitats fall cleanly into this binary classification, with each site either entirely within or entirely outside a Natura 2000 Special Area of Conservation. Surrounding land use is included separately in the models (forest cover, arable cover, urban cover, road type and road distance), preserving an interpretive separation between legal protection and land-cover gradients.

Acoustic recording protocol

Audio recordings were obtained using the six SongMeter Micro devices (Wildlife Acoustics Inc., Maynard, USA) and one AudioMoth 1.1.0 device (Hill et al. 2019) enriched by original waterproof case. To ensure consistency, all devices were configured with identical settings; .wav format, at a bit depth of 16-bits, with a 48 kHz sampling rate and mid gain (mid gain in AudioMoth, 12 dB in SongMeter Micro), spectrograms of test recordings were inspected. AudioMoth device was deployed at the same site throughout the research (Site 7 Kasarna). Additionally, throughout the year, we conducted random sensitivity tests on the microphones across all devices via test recording of playback and screening the spectrograms. We used songs of two common passerine species — Common Chaffinch (Fringilla coelebs) and Eurasian Wren (Troglodytes troglodytes) — to represent typical mid-frequency bird vocalizations. Recordings were played back at a calibrated sound pressure level of 70 dB SPL at 1 m (approximating the sound level of a nearby singing passerine within a natural dawn chorus.). Both recorders were placed at distances of 3, 5, and 10 m from the speaker, with three replicates at each distance for each device. A short 1 kHz calibration tone (250 ms) preceded each playback to allow post-processing alignment and amplitude normalization. The test was conducted in a quiet outdoor environment to minimize background noise (see spectrograms and more details in Annexes, Supplementary Figure S1).

Acoustic recordings were obtained in 10-minute files, with one recording taken every 30 min (10 min on, 20 min off) throughout the entire 24-hour daily cycle. This process was repeated for 7–10 consecutive days each month, spanning from March 2022 to February 2023. Due to researcher absence, recording devices could not be maintained in June and August. The AudioMoth at Site 7 Kasárna resumed recording in August but encountered technical issues in October and December, producing uneven audio files; data from these months were therefore excluded from analysis. All recorders were mounted on tree trunks at the height of 2.0–3.0 m heading along the bank of the oxbow lakes. We aimed to capture sounds from both terrestrial (birds, insects, mammals) and aquatic environments (anurans), and avoided dense vegetation within several meters in front of the microphone to ensure clear sound propagation.

Acoustic indices

To quantify patterns of biophonic activity, we selected two complementary acoustic indices: the Mid-Frequency Cover (MFC) and the Acoustic Complexity Index (ACI). MFC has demonstrated sensitivity to vocal activity across a wide range of bird species (Yip et al. 2021; Allen-Ankins et al. 2023) and has shown predictive power for species richness when applied to new datasets (Haupert et al. 2025). ACI was developed to detect variations in biophonic sounds, which typically fluctuate in intensity, while filtering out persistent anthropophonic noise (Pieretti et al. 2011). It has been widely used as a measure of biophony in environmental recordings, and has shown as one of the most reliable index to correlate with species diversity (Alcocer et al. 2022). Both indices were generated from the audio for the entire monitoring periods at a 1-min resolution using QUT Ecoacoustics Analysis Programs (Towsey 2017; Version v21.7.0.4), with default analysis settings (512-sample frames, Hamming window) applied to the 48 kHz recordings.

Mid-Frequency Cover (MFC; Towsey 2017) is a binary cover index. For each 1-min segment, the noise-reduced spectrogram is partitioned into time–frequency cells, and MFC is computed as the proportion of cells in the 1–8 kHz band whose amplitude exceeds 3 dB above the estimated background level. Values therefore range from 0 (no mid-frequency sound above threshold) to 1 (all cells above threshold), and are most sensitive to the temporal–spectral occupancy of the mid-frequency band by acoustic events. The selected frequency range was chosen to capture the majority of bird vocalisations while reducing the influence of low-frequency anthropogenic noise, which typically dominates below 1 kHz.

The Acoustic Complexity Index (ACI; Pieretti et al. 2011), by contrast, is computed from the absolute amplitude differences between consecutive frames within narrow frequency bins, and is most sensitive to the rapid intensity variation characteristic of biophony, while being relatively — though not entirely—impervious to constant background noise (Bradfer-Lawrence et al. 2023). Conceptually, MFC quantifies how much of the mid-frequency soundscape is occupied by acoustic events, broadly reflecting the amount of acoustic energy within the defined range. Whereas ACI captures short-term amplitude variability across frequency bins and is therefore most closely related to the variability of sound intensity – an aspect often linked to acoustic diversity. ACI can also be influenced both by biophony and by intermittent geophony or anthropophony in the same band (Bradfer‐Lawrence et al. 2024). The two indices thus highlight different aspects of the soundscape, though they are often correlated. To exclude recordings dominated by heavy geophony, we used the ‘High Amplitude’ and ‘Clipping Index’ metrics to identify files affected by loud wind and rain, excluding all recordings with values > 0 from further analysis (following the first filtering step of Müller et al. 2022). To verify the filter, we additionally inspected a random subset of 20 excluded recordings, all of which were indeed dominated by geophony (wind or rain), confirming low commission error of the filter. Because this amplitude-based step targets recordings dominated by heavy weather events, it does not necessarily remove lighter geophonic events such as intermittent rain drops falling from vegetation or short wind gusts, which may have remained in the analysed dataset; their possible influence on the indices is addressed in the Discussion. For each remaining 10-min recording and each acoustic index, the 1-min values were then averaged to obtain a single recording-level estimate (Bradfer‐Lawrence et al. 2023).

Aural classification

Following the protocol developed by Gasc et al. (2018) and modified by Müller et al. (2022), we applied a stratified random sampling scheme to select a subset of our recordings for detailed analysis. Specifically, we analysed 10 one-minute recordings per site (7), per month (10), and per day phase (2 - dawn and dusk). For each month and site, we randomly selected two dates, and within each date, we extracted five recordings during dawn (30 min before to 2 h after sunrise) and five during dusk (2 h before to 30 min after sunset). All random selection were conducted in R 4.4.0 (R Core Team, 2024). This resulted in a total of 1380 1-min audio files.

This design allowed us to cover a broad range of vocal activity timing, which varies among species and seasons in temperate zones. Each 1-minute sample represented a larger 30-minute period, effectively capturing intra-period acoustic diversity and vocalization peaks. The same temporal structure was applied across all sites (except Site 6 Kasarna, see below), thereby avoiding weather-related biases. Importantly, this “two-day-per-month” sampling strategy mirrors standard practices in breeding bird surveys, where a limited number of targeted visits per season are considered sufficient to detect the majority of species present (Bibby 2000; Sutherland et al. 2004). Our protocol spans both the breeding and non-breeding periods, offering a representative cross-section of the annual acoustic activity of the bird community at each site. This increases the interpretative power of the dataset and supports its use as a proxy for relative species richness and biophonic activity across habitat types and time. Species accumulation curves were generated to evaluate the relationship between sampling effort and cumulative species richness at each site. This approach allowed us to assess the completeness of the sampling, compare richness patterns among sites, and identify whether additional sampling would likely yield new species records. Curves were calculated in R (version 4.4.0) using the ‘vegan’ package (version 2.6-6) and incidence-based presence–absence data across monthly samples, with species order randomised through multiple permutations to produce smoothed estimates (Fig. 6).

To classify bird species manually, the co-author, DA, assessed the same audio files by listening and visually analyzing spectrograms using Raven Pro 1.6.5. (Raven Sound Software, 2024). Further the files were categorized into three soundscape types: biophony, geophony, and anthropophony (following Pijanowski et al. 2011). Some files contained multiple categories. The animal group (birds, anurans), type and aural intensity of selected subcategories within geophony and anthropophony were recorded but were not processed in this work (Supplementary Table S2).

Statistical analyses

In order to analyze the effect of site characteristics and seasonality on bird species richness, two-sample t-tests and one- and two-factor analysis of variance and correlation analysis were employed in the statistical analyses, see for example (Devore 2000). The relationship between bird species richness and acoustic indices was evaluated through the use of Spearman correlation analysis methods, as outlined, for example, in Devore (2000).

For this purpose acoustic index values were aggregated in the following way: for each site, 10-minute indices values were aggregated to daily means, which were then further aggregated to a single monthly mean value. For dawn and dusk analyses, only recordings within the predefined time windows (described above; relative to sunrise or sunset, adjusted for seasonal variation in day length) were included in the aggregation process. Spearman correlation was calculated in R (version 4.3.2).

Individual acoustic indices were modeled using generalized additive models (GAM). Categorical variables (site, month, and site conservation status) and continuous variables (time t) were included in the model. The model can be written in the form

\(\:i=1,\:2,\dots\:,\:7,\:j=1,\:2,\dots\:,\:12,\:k=1,\:2,\:t\in\:\left[\text{0,24}\right),\:\)where\(\:\:{\mu\:}_{ijk}\) is a mean of an acoustic index, \(\:g\) is a link function, \(\:c,\:{\alpha\:}_{i},\:{\beta\:}_{j},\:{{\gamma\:}_{k},\:\lambda\:}_{ij}\) are parameters. The constant c in model (1) denotes the so-called grand mean. The parameter \(\:{\alpha\:}_{i}\) is the effect of the site, \(\:{\beta\:}_{j}\) is the effect of the month, \(\:{\gamma\:}_{k}\) is the effect of the site conservation status, \(\:{\lambda\:}_{ij}\) is the interaction between month and site conservation status, and \(\:{s}_{jk}\left(t\right)\) are cyclic cubic regression splines for month and site conservation status for time t, see e.g. (Hastie 2009), (Wood 2017). GAM model parameter estimates were calculated in R (version 4.3.2) using the mgcv package (version 1.9-0).

Effects of seasonality on diurnal and annual patterns of acoustic indices

The parametric coefficients for individual months and specific sites, derived from Generalized Additive Models (GAMs), revealed significant temporal variation across all modelled acoustic indices. Monthly values were compared against the overall mean (intercept) to assess seasonal trends. In general, both ACI and MFC exhibited elevated values during the spring period, followed by declines across the rest of the year, with minor temporal shifts among indices. Specifically, ACI values increased from February to May, and MFC from April to July, with MFC also displaying a secondary peak in October (Figs. 3 and 4, Tables S2 and S3); ACI additionally showed a slight increase in December, particularly during nighttime hours. The magnitude of monthly contrasts was approximately an order of magnitude larger for MFC than for ACI (e.g., May estimates on the log scale: +0.814 vs. +0.027; Tables S2–S3).

Diurnal patterns of both indices were characterised by a dominant morning peak (approximately 6–9 h), most pronounced from late winter through spring (Feb–Apr), and a weaker secondary afternoon/evening rise apparent in some months — particularly Oct–Dec for MFC (Figs. 3 and 4). The timing of the morning peak shifted in accordance with seasonal changes in sunrise. Across the annual cycle, the smallest differences between nighttime and daytime values were recorded in the summer (June–September) and winter months (December–January). The diurnal smooths for MFC remained highly structured year-round (edf typically 6–10, p < 0.001 across nearly all month × site-conservation combinations; Table S3), whereas ACI’s diurnal structure weakened markedly outside the breeding season, becoming nearly flat in July and September (edf ≈ 1–2.5; the July smooth for [unprotected] sites was non-significant, p = 0.174; Table S2). Overall, the GAM for MFC explained 53.1% of deviance compared with 36% for ACI (Tables S2 and S3), indicating that MFC captured a substantially larger fraction of the soundscape’s spatio-temporal structure.

Soundscape and species composition

Aural inspection revealed that biophony was present in nearly all dawn recordings across sites year-round, except at Site 7, which showed decreases in January and March (Fig. 5). During dusk, biophony occurrence was less consistent, with lowest detections generally in January and, for Sites 5 and 6, also in September. Anthrophony was detected almost continuously during dawn at most sites, except at Site 6 in February, March, and May; dusk detections were more variable, with Site 4 showing stable proportions and Site 6 largely lacking these sounds from spring to autumn. Geophony was more common during dusk than dawn, mainly light to moderate rain and wind, with heavy-weather recordings excluded during preprocessing. Highest geophony levels occurred in spring and summer. Most biophonic events consisted of bird sounds. Protected sites generally maintained higher and more stable biophony levels across seasons, whereas unprotected sites showed greater variability and more frequent reductions in biological sound components.

The dataset comprised 88 bird species with a total of 1,229 recorded presences across months and sites (see Supplementary Figure S4). Species richness peaked in May (52 species; 159 presences), was already high in March–April, and declined toward autumn and winter. The most frequently recorded species were Great Tit (Parus major), Blue Tit (Cyanistes caeruleus), Common Blackbird (Turdus merula), Great Spotted Woodpecker (Dendrocopos major), Common Chaffinch (Fringilla coelebs), and Hawfinch (Coccothraustes coccothraustes). Eighteen species occurred exclusively at protected sites, including Marsh Warbler (Acrocephalus palustris), Tree Pipit (Anthus trivialis), Short-toed Treecreeper (Certhia brachydactyla), and Stock Dove (Columba oenas). Thirty-two species were present only in the warm part of the year (no records in Dec–Feb), a period dominated by insectivores and mixed-feeding species such as Marsh Warbler, Eurasian Skylark (Alauda arvensis), Tree Pipit, Common Swift (Apus apus), and Short-toed Treecreeper. In February, 45 species were recorded, 43 of which persisted into spring/summer (Mar–May, Jul, Sep); only Coal Tit (Periparus ater) and Redwing (Turdus iliacus) were recorded exclusively in February (in the first half of the year), when omnivores and granivores predominated. In October, 37 species were recorded relative to the previous month period (May, Jul, Sep), with eight unique to October—Rook (Corvus frugilegus), Brambling (Fringilla montifringilla), Coal Tit, Grey-headed Woodpecker (Picus canus), Eurasian Bullfinch (Pyrrhula pyrrhula), Common Firecrest (Regulus ignicapilla), Goldcrest (Regulus regulus), and Eurasian Siskin (Spinus spinus)—while the remaining 29 overlapped with the breeding/post-breeding assemblage. Compared to September, October site-level presences increased at Site1 (+ 7), Site2 (+ 6), Site4 (+ 3), and Site5 (+ 7), and decreased slightly at Site3 (− 2) and Site6 (− 1) (Fig. 6 and Supplementary Figure S4 ).

Effects of site conservation on temporal trends of acoustic indices

The GAMs revealed associations between site conservation and both acoustic indices, although the magnitude and statistical significance of the conservation effect differed between them. The factor conservation explained relatively small additional variance — approximately 1.6% in ACI and 2% in MFC — and the main effect of conservation status was significant only for ACI (parametric estimate ± 0.019, p = 0.035; Table S2), not for MFC (± 0.132, p = 0.272; Table S3). Despite this, interaction terms with month were significant for both indices, indicating that conservation status modulated the seasonal expression of acoustic activity. In MFC, conservation appeared to slightly enhance the dawn and dusk peaks, particularly from autumn through winter, and both indices showed consistently high values throughout the year at NATURA 2000 Sites 1 and 4 (Figs. 3 and 4).

Effect of site characteristics and seasonality on bird species richness

Bird vocal species richness differed significantly among sites among sites (ANOVA, F-test 2.381, p < 0.05; Fig. 7). Linear models indicated that mean species richness was significantly lower than the unweighted mean at Site 3 (p < 0.05) and significantly higher at Site 5 (p < 0.05), with a similar positive but non-significant trend at Site 1 (p = 0.092). T-tests revealed significant difference in number of vocalizing bird species between dawn and dusk (t = 4.854, p-value < 0.001; Fig. 7A). When daypart was considered, richness at Site 3 was again significantly lower and at Site 5 significantly higher than the unweighted mean (both p < 0.05), with marginal positive deviation at Site 1 (p = 0.066) and marginal negative deviation at Site 7 (p = 0.087; Fig. 7B). Species richness also varied strongly among months (ANOVA, F = 7.900, p < 0.001; Fig. 7C). Species richness was further associated with site characteristics and landscape context. Significant relationships were detected for site conservation status (T-test − 3.402, p < 0.001; Fig. 8A) and road type (T-test 2.166, p < 0.05; Fig. 8B).

Across the annual dataset, species richness was most strongly and positively correlated with distance to roads, while urban area cover showed a strong negative relationship (Table 2 and S4). These patterns were particularly pronounced during spring and autumn migratory periods, when road distance exhibited high positive correlations with species richness (spring: ρ = 0.90, p_adj_BH = 0.022; autumn: ρ = 0.85, p_adj_BH = 0.065), and urban area cover showed correspondingly strong negative correlations (spring: ρ = − 0.74, autumn: ρ = − 0.78). Spearman correlations between bird species richness and monthly aggregated index values and were consistently stronger for MFC than for ACI (Table 3): overall ρ = 0.639 vs. 0.418, dawn ρ = 0.570 vs. 0.273, and dusk ρ = 0.422 vs. 0.226 (the latter only marginally significant for ACI).

Intra-annual and diurnal patterns of vocal activity and acoustic indices

The intra-annual monitoring of riparian oxbow lake soundscapes revealed pronounced seasonal and diurnal structuring of temperate acoustic environments. The substantially higher deviance explained by the MFC model (53.1%) compared with ACI (36%) and the order-of-magnitude stronger seasonal contrasts in MFC indicate that this index captured the phenological dynamics of the soundscape more effectively. Spring maxima in both indices coincided with the peak of avian species richness across all sites (Fig. 2) and with the dawn chorus driven by breeding-season vocal activity (Slagsvold 1977). The seasonal shift of the morning peak in line with sunrise further highlights the sensitivity of acoustic indices to environmental cycles (Müller et al. 2024). The year-round persistence of structured diurnal patterns in MFC, contrasted with the near-flat ACI smooths during the post-breeding period (July–September), suggests that MFC remains responsive to soundscape structure even when avian vocal activity declines, whereas ACI is more narrowly tied to the breeding season. Beyond their differing capacity to capture overall soundscape structure, the two indices also differed substantially in how closely they tracked bird species richness, with MFC showing consistently stronger correlations than ACI across overall, dawn, and dusk subsets (Table 3). This differential performance likely reflects fundamental differences in how the indices are computed rather than a conceptual hierarchy between them. ACI, as calculated here, integrates rapid amplitude variability across the full spectrum and therefore captures contributions from a broad range of sound sources, including anurans, insects, and low-frequency components of geophony and anthropophony (Pieretti et al. 2011; Bradfer-Lawrence et al. 2023). MFC, in contrast, is restricted to the 1–8 kHz band where most bird vocalisations are concentrated (Towsey 2017), and as a temporal–spectral cover measure scales more directly with the cumulative duration and overlap of vocal events within that band. In simple terms, ACI characterises how complex and dynamic the sound activity is over time, whereas MFC characterises where the energy is concentrated in the frequency spectrum. When acoustic patterns are dominated by diverse and dynamic bird vocalisations, the two indices should covary; when they deviate, this can indicate either steady but non-dynamic bird sound (high MFC, low ACI) or dynamic acoustic activity originating outside the mid-frequency band (low MFC, high ACI). This is important because acoustic indices do not respond to species richness directly, but rather to the acoustic activity of the assemblage — the combined effect of the number of signalling species, the number of signalling individuals, and species-specific calling rates and song characteristics (Alcocer et al. 2022; Bradfer‐Lawrence et al. 2023). The frequency-band specificity of MFC therefore provides a parsimonious explanation for its consistently stronger association with bird species richness.

A notable seasonal shift occurred in October, when bird species richness and MFC values increased relative to September (Figs. 3, 4 and 2; MFC October estimate + 0.044, p < 0.001, Table S3); ACI showed only a modest change at this time (Table S2). This shift reflects the onset of autumn migration and the role of riparian habitats as resource-rich stopovers within intensively used cultural landscapes (Bonter et al. 2009). Species composition data support this interpretation, eight species were unique to October, many of them granivores or mixed-feeders such as Brambling (Fringilla montifringilla), Common Firecrest (Regulus ignicapilla), and Goldcrest (Regulus regulus), that are typical autumn migrants exploiting seed-rich habitats after insect availability declines. The simultaneous increase in site-level presences at most locations compared to September suggests intensified stopover use across the network. This seasonal shift was also reflected in the reduced contrast between dawn and dusk acoustic peaks, consistent with the simpler and less synchronised migration calls typical of this period (Gayk and Mennill 2023).

Beyond the spring peak, several phenological signals deserve attention. In February, unexpectedly high species richness and elevated acoustic index values likely reflected early-season activity of resident and wintering species, combined with the arrival or passage of early migrants. Of the 45 species recorded that month, only Coal Tit (Periparus ater) and Redwing (Turdus iliacus) were exclusive to February, indicating that most belonged to assemblages continuing into spring. Many were omnivores or granivores capable of exploiting winter food resources, which, together with favourable microclimatic conditions in riparian zones (Scheffers et al. 2014; Ellis 2020), may explain the elevated activity. Guild-level dynamics closely tracked seasonal prey availability, with insectivores peaking in spring and summer and omnivores buffering winter communities, patterns that aligned with seasonal shifts in the acoustic indices. These observations highlight the sensitivity of acoustic indices to detecting phenological shifts (Buxton et al. 2016) and reinforce the role of riparian habitats as both early-activity hotspots and critical migratory stopovers. To assess the ecological realism of our acoustic detections, we qualitatively compared the species inventory aurally identified at Site 6 (Kanada) — from the 200 one-minute recordings constituting the aural-classification subset for this site (10 recordings × 10 months × 2 dayparts) — against records from the Czech Society for Ornithology database (avif.birds.cz; Česká společnost ornitologická 2026) for the surrounding Kněžpolský les area, accumulated over the past decade (Supplementary Table S5). Site 6 was selected because it is the only one of our seven sites for which sufficient long-term avif records were available. The 10-year avif window was used deliberately to capture a robust species pool for the area; because avif is an opportunistic citizen-science dataset rather than systematic monitoring, a shorter window would not provide a meaningful match in effort, and the comparison is therefore intended only as a qualitative cross-check on the ecological realism of the acoustic identifications. Of the 53 species reported in avif, 33 (62%) were also identified in the aural subset for Site 6, including most dominant woodland and riparian taxa; species reported only in avif were predominantly raptors and soaring birds, consistent with known limitations of passive acoustic monitoring (Hoefer et al. 2023). PAM in turn identified 16 species not present in the avif records for this locality — including nocturnal, crepuscular, and wetland-associated taxa — likely underrepresented in opportunistic visual surveys(Sidie-Slettedahl et al. 2015; Bobay et al. 2018), supporting the treatment of PAM-derived and observation-based datasets as complementary rather than redundant.

Effect of site characteristics and conservation

The acoustic and richness patterns observed across sites reflect the combined influence of protection status, landscape context, and local habitat structure, which we discuss in turn. MFC and ACI differed in how they separated conservation from non-conservation sites. The main effect of site conservation was statistically significant for ACI (parametric estimate ± 0.019, p = 0.035; Table S2) but not for MFC (± 0.132, p = 0.272; Table S3), and conservation status as such accounted for only a small fraction of the total variance in either index (~ 1.6% in ACI, ~ 2% in MFC). Visual inspection of the diurnal curves (Figs. 3 and 4) nevertheless indicates a clearer grouping by conservation status in ACI than in MFC, where curves for conservation (Sites 1, 4, 5, 6) and non-conservation (Sites 2, 3, 7) sites frequently overlap. The relatively modest overall variance attributable to conservation status, combined with the contrasting separation between the two indices, is consistent with sites differing less in the overall amount of avian vocal activity — which MFC primarily reflects — than in the variability of their acoustic patterns, which ACI emphasises. One plausible interpretation is that conservation sites host bird communities with more dynamic song characteristics, potentially reflecting community dissimilarity rather than richness per se. However, the acoustic differences between conservation and non-conservation sites likely arise from a combination of community composition, landscape context, and exposure to anthropogenic sound, which cannot be fully disentangled with the present design; dedicated community-level analyses, for example based on dissimilarity metrics or song-trait composition, would be needed to confirm the role of community structure directly. More broadly, acoustic indices integrate both the number of vocalising species and their abundance and calling activity, so higher index values at protected sites may reflect not only more species but also greater abundance and vocal output of the species present (Alcocer et al. 2022; Bradfer-Lawrence et al. 2023).

Beyond the conservation-effect signal discussed above, landscape context strongly shaped bird species richness across our sites. Richness, particularly during migratory periods, increased with distance from major transport infrastructure and decreased with surrounding urban cover (Table S4, Fig. 2), consistent with evidence that anthropogenic noise alters avian acoustic behaviour and lowers diversity (Arroyo-Solís et al. 2013; McClure et al. 2013). Urban areas may sustain relatively high bird numbers through food availability (Pautasso et al. 2011), but traffic noise can mask vocalisations, hinder predator detection, and reduce fitness (Barber et al. 2010; Ortega 2012). These landscape effects were weaker during breeding and wintering periods, when more stable resident communities likely reflect local habitat quality rather than broader landscape gradients. Higher cumulative species richness at several protected sites is therefore best understood as the joint outcome of legal protection and favourable landscape context (Shaver et al. 2022), rather than of either factor alone.

Within this landscape framing, ecotonal conditions emerged as an important driver of local richness. The small protected Site 1 and the protected Site 5, despite their contrasting landscape context, both supported high species richness; their shared feature was strong connectivity with riparian forest edges and structurally diverse vegetation buffers that concentrate resources and promote avian activity (Andrén 1994; Terraube et al. 2016). Similar ecotonal effects have been reported in restored oxbows (Shaver et al. 2022) and riparian soundscapes (Budka et al. 2023), suggesting that well-developed edges can rival or exceed continuous forest habitats in supporting diverse bird assemblages. In contrast, the protected Site 6, located within a large forest interior, showed lower detected richness, likely reflecting spatial dispersion of vocal activity and reduced detectability (Zanette et al. 2000; Beason et al. 2023).

Several methodological considerations should be acknowledged. First, the dataset was limited to seven sites and used two recorder models: the AudioMoth at the unprotected Site 7, and SongMeter Micro at the remaining sites. The AudioMoth exhibited lower overall amplitude and greater internal noise (Supplementary Figure S1), which may have contributed to the consistently lower MFC values recorded at Site 7. Complexity-based indices and aural species detections are likely less affected, and the broad consistency of ecological patterns across devices suggests that the main conclusions are robust; nevertheless, site-level contrasts in amplitude-sensitive indices should be interpreted with this technical variation in mind. Second, species accumulation curves did not reach an asymptote at any site, indicating that further sampling would likely yield additional species records. Because the curves had not levelled off, the most interpretable difference between site categories was in the overall species totals reached rather than in the rate of accumulation: several protected sites supported larger cumulative species counts than most unprotected ones, reinforcing the role of conservation status alongside habitat heterogeneity, surrounding land use, and seasonal turnover in shaping diversity. Third, our amplitude-based screening of recordings dominated by heavy weather addresses commission errors of the geophony filter (Methods), but does not fully prevent omission errors. Both indices are potentially affected by geophony to varying degrees, and the exact magnitude of these effects is difficult to disentangle from our data. Lighter geophonic events such as intermittent wind or rain dripping from vegetation may have remained in the dataset and can elevate ACI in the absence of biophonic activity (Sánchez-Giraldo et al. 2020; Bradfer‐Lawrence et al. 2023, Darras et al. 2024). Mechanistically, the restriction of MFC to the 1–8 kHz band would be expected to make it less sensitive to broadband low-frequency geophonic energy than ACI, which integrates across the full spectrum. However, MFC has been comparatively less tested in this respect. Our recommendation of MFC as the more reliable biophony-tracking index for temperate riparian soundscapes therefore rests primarily on its stronger empirical association with bird species richness in this study (Table 3), rather than on a tested difference in geophony sensitivity.

Taken together, our results add to the growing recognition that no single acoustic index serves as a one-to-one proxy for species richness; rather, indices reflect different facets of acoustic activity, and their joint interpretation—with explicit attention to the distinction between richness, abundance, and acoustic activity—is essential (Alcocer et al. 2022; Allen-Ankins et al. 2023; Bradfer-Lawrence et al. 2023; Hoefer et al. 2023). For temperate riparian soundscapes of the type studied here, we recommend MFC as the more reliable biophony-tracking index, with ACI as a complementary but more context-sensitive metric, particularly when paired with targeted aural inspection. The consistent diel and seasonal patterns recovered across the annual cycle indicate that short-term perturbations did not obscure the broader biophonic structure, supporting the value of long-term passive acoustic monitoring for separating transient disturbances from ecological signals. From a conservation perspective, our findings support the effectiveness of formal protection, and restored or well-buffered systems provide refuge, promote seasonal turnover, and support sensitive bird taxa (Shaver et al. 2022). Formal protection—as in NATURA 2000 sites—together with complementary active management of small wetland systems, emerges as a cost-effective approach for biodiversity conservation in human-dominated cultural landscapes. Ecoacoustic monitoring offers a scalable approach for tracking the seasonal and diurnal dynamics of vocal bird communities across the annual cycle, providing an indirect window onto habitat quality.

Year-round ecoacoustic monitoring revealed that riparian oxbow lakes are highly dynamic soundscape systems shaped by strong seasonal and diurnal turnover in vocal activity. Peaks in acoustic indices and bird species richness reflected breeding phenology, migration pulses, and shifting guild composition across the annual cycle, with each phase contributing taxa absent in other seasons—illustrating that capturing the full community using these habitats requires sampling well beyond the breeding season alone, and highlighting the role of oxbows as ecological refugia and migratory stopovers in human-dominated landscapes. Landscape context modulated these patterns, with lower urbanisation and greater distance from transport infrastructure supporting richer assemblages, particularly during migration. Importantly, oxbow lakes with formal legal protection consistently exhibited higher acoustic activity and greater cumulative bird species richness than unprotected sites, indicating that conservation status translates into measurable acoustic and ecological benefits even at the scale of small wetland systems.

Methodologically, among the tested indices, MFC tracked bird species richness more closely than ACI, supporting temporal–spectral cover metrics as a reliable measure of temperate riparian biophony, while ACI remains more sensitive to local acoustic context. The study reinforces that acoustic indices should be interpreted as measures of acoustic activity—the combined product of richness, abundance, and calling behaviour — rather than direct proxies of species richness alone. Combined with targeted manual validation, passive acoustic monitoring captured both early-season biological activation and autumn migratory pulses, demonstrating that ecoacoustics, integrated with established field methods, offers a scalable tool for long-term biodiversity monitoring, conservation planning, and restoration assessment in riparian environments.

Authors

1. M. Adam
2. D. Adamová-Ježová
3. J. Neubauer
4. S. Müller

Cite this article

Adam, M., Adamová-Ježová, D., Neubauer, J. et al. Riparian soundscape dynamics of Central European oxbow lakes: insights from year-round ecoacoustic monitoring. Biodivers Conserv 35, 195 (2026). https://doi.org/10.1007/s10531-026-03393-x

Download citation

• Received: 19 August 2025

• Revised: 16 May 2026

• Accepted: 11 June 2026

• Published: 25 June 2026

• Version of record: 25 June 2026

• DOI: https://doi.org/10.1007/s10531-026-03393-x

Keywords