The study was conducted in and around Namdapha Tiger Reserve (NTR), a 1985 km² protected area in Arunachal Pradesh, India (Fig. 1), situated within the Indo-Burma biodiversity hotspot (Myers et al. 2000). This region is home to extensive rainforests (Champion and Seth 1968) and is considered the northernmost lowland tropical wet evergreen forest in the world (Proctor et al. 1998). Due to its wide elevation range (200–4,571 m), the forest types transition from wet evergreen forests to wet temperate forests with increasing altitude (Champion and Seth 1968). The park’s habitat heterogeneity supports highly diverse species assemblages, including several endemic species. Previous amphibian surveys in this region employed observer-based survey methods and did not incorporate passive acoustic monitoring (Sarkar and Sanyal 1984; Dinesh and Radhakrishnan 2019).

The study area map was generated using the digital elevation model (DEM, 30 m resolution) from the NASA Shuttle Radar Topography Mission (SRTM) (NASA JPL 2013). This DEM was used to generate a terrain ruggedness layer using the Terrain Ruggedness Index (TRI) tool in QGIS (QGIS Development Team 2024).

Fourteen sampling locations were selected in the tropical evergreen forest zone of the study area for the survey (Fig. 1), based on existing literature (Sarkar and Sanyal 1984; Dinesh and Radhakrishnan 2019), previous exploration surveys conducted in 2022, and information provided by local people. Each sampling location (site) corresponded to one of the following four habitats: (1) lake/pond, (2) river/stream, (3) roadside water puddle, and (4) swamps within the forest. Data were collected through PAM and VES methods at the beginning of the wet season (May–June 2023) in Northeast India to avoid the risk of floods and landslides in the following months.

Song Meter Micro units (Wildlife Acoustics, Inc.) were used for passive acoustic monitoring. Each PAM unit was powered by three AA alkaline batteries, and silica bags were placed inside to prevent moisture accumulation. Digital recordings were stored on microSD cards inserted in the units. The devices were configured using the Song Meter Configurator mobile app (v. 1.5.15). The sampling rate was set to 44.1 kHz, gain was set to 18 dB, and individual recordings were saved as 60-min files. PAM units were configured to record from dusk to midnight (18:00 to 23:59 h), coinciding with the peak activity period of amphibians. At each of the 14 sampling sites, one PAM unit was installed to record continuously for 6 h over 1 day. Each device was securely strapped to a tree or plant and positioned approximately 2 m above the ground (Walls et al. 2014).

At each ARU-installed location, a nocturnal visual encounter survey (VES) (Scott et al. 1994) was conducted between 19:00 and 21:00 h on the day following acoustic data collection. Three surveyors searched for anurans at each site for a fixed duration of 2 h using flashlights. The VES sampling effort at each site was 6 person-h (3 persons × 2 h), which is consistent with the PAM sampling effort of 6 h (1 unit × 6 h).

Raven Pro (v. 1.6) was used for analyzing the sound files retrieved from PAM units. Sampling effort was standardized to 10 min for each hour of recording (Melo et al. 2021; Ribeiro Jr et al. 2022). The 10-min continuous segment within each h was randomly selected. Thus, for each site, 60 min (6 h × 10 min/h) of call recording was considered the sample for analysis. Analysis involved listening to the recordings and visually scanning the spectrogram in Raven Pro, with frame length set to 60 s (Wimmer et al. 2013; D’Anunciação et al. 2022). The spectrum function was set to a Hanning window of 256 samples and a DFT size of 1024 samples. Two team members with adequate knowledge of anuran calls listened to the recordings and identified the species. The identified calls were annotated, and the dominant frequency was measured using peak frequency measurement in Raven’s selection spectrum function (Bernardes et al. 2015). Dominant frequency is defined as the frequency at which the maximum power is observed in a call selection.

Gamma species richness was estimated for VES and PAM methods separately using a non-parametric bootstrap approach (Efron and Tibshirani 1994), wherein 14 sites were resampled with replacement 1000 times, and the number of unique species in each replicate was recorded. The resulting distribution of the estimated richness was used to calculate the 95% confidence interval (Efron and Tibshirani 1994; Chiu et al. 2014). Given that the sampling sites were independent and the sampling effort was consistent, this resampling-based method is flexible and widely applied in ecological studies (Gotelli and Colwell 2001). The bootstrap-derived species richness estimates of the two methods were compared using the non-parametric Mann–Whitney U test (Mann and Whitney 1947).

Additionally, Jaccard’s index was used to measure the similarity between the set of amphibian species detected by PAM and VES. The species were grouped based on ecological traits to investigate the method-specific detection patterns (Davis et al. 2025). The analyzed traits were (1) vertically stratified habitat—fossorial, stream, terrestrial, low-vegetation, and arboreal; (2) body size using snout–vent length (SVL)—small (≤ 30 mm), medium (31–60 mm), and large (> 60 mm); and (3) dominant call frequency—low (< 2 kHz), mid (2–4.5 kHz), and high (> 4.5 kHz). For each trait group, asymptotic species richness was estimated using the incidence-based estimators implemented in iNEXT, which modeled the expected richness and provided the 95% confidence intervals. Given the lack of published literature on traits for a few species, the literature-based values were supplemented with field data from this study. The referred literature for species traits is listed in Table 1. SVL values correspond to the average snout–vent length of males. In cases where a species was detected exclusively through PAM but could not be reliably identified, the site was revisited to locate the calling individuals. Their SVL was measured (n ≥ 3), and associated habitat characteristics were recorded. Dominant call frequency corresponds to the mean dominant frequency of the species. For species exhibiting multiple call types, the call type with the highest mean dominant frequency was used.

To examine species richness and sampling adequacy, species accumulation curves (SAC) were constructed for (1) PAM, (2) VES, and (3) PAM+VES methods, wherein the number of sampled sites was chosen as the effort. R software (v. 4.1.3) and the iNEXT package (Hsieh et al. 2016) were used to run the above analysis. To estimate the effort required to detect all species, species detections were extrapolated to three times the minimum observed sample size (Hsieh et al. 2016).

The data that support the findings of this study are deposited in the Figshare data repository at https://doi.org/10.6084/m9.figshare.32323527 (Rajiv 2026).

After excluding corrupt files and those with excessive noise from rain, 3,360 min of acoustic recordings were obtained. From 720 min of subsampled data (10 min per h), a total of 19 amphibian species belonging to 17 genera and six families (Frost 2024) were identified (Table 1). Visual encounter surveys also detected 19 species of amphibians belonging to 17 genera and seven families (Frost 2024) (Table 1). Both PAM and VES detected 11 endemic amphibian species (Table 1, Fig. 2). The bootstrap-derived 95% confidence interval for species richness ranged between 14 and 19 species for both PAM and VES methods (Fig. 3). A clear overlap was observed in the estimated richness confidence intervals, indicating similarity in the output of the two methods. The Wilcoxon rank-sum test confirmed that there was no significant difference in regional species richness between methods (W = 494174, p > 0.1).

However, a marginal difference was observed in the composition of species detected by the two methods. The Jaccard similarity coefficient, or Jaccard index, was 0.652, which implies that 65.2% of the total species were recorded by both methods, while 17.4% of species were unique to each method. Fig. 4 displays the species detected by PAM and VES based on ecological traits. The results show that PAM was effective in detecting arboreal and fossorial anurans, whereas VES detected a greater number of species inhabiting streams. In terrestrial habitats, although VES detected two additional species, the estimated richness remained similar between methods, with overlapping confidence intervals. Across body-size classes, both methods revealed nearly identical species counts, with PAM detecting only one additional small-bodied species (SVL < 30 mm), Raorchestes nasuta. Across call-frequency groups, species richness patterns differed modestly between the two methods (Fig. 4). Among high-frequency callers, VES recorded two species, whereas PAM detected only one species. In the low-frequency group, both methods detected six species, but the estimated richness was higher for VES than for PAM. For mid-frequency callers, observed and estimated richness values were similar between methods, with broadly overlapping confidence intervals. Limnonectes sp. could not be identified to the species level, and hence it was not possible to retrieve its representative dominant frequency value from the literature. Consequently, Limnonectes sp. was excluded from analyses examining the relationship between dominant frequency and species richness.

Interestingly, the upper limit of the bootstrap confidence interval of richness for both VES and PAM methods was not greater than the observed richness (Fig. 3). This disparity was investigated by evaluating sampling adequacy. Species accumulation curves (SAC) were constructed based on the species detection history across sampling locations. For both methods, a positive but decelerating trend in cumulative species richness was observed with increasing sampling sites, indicating that new species continue to be detected, albeit at a slower rate (Fig. 5). The rate of species accumulation at the tail end of the rarefaction curve—calculated by measuring the slope of the last three points—was 0.46 for PAM and 0.46 for VES. This indicates that, on average, 0.46 new species were added with each additional site.

When both survey methods were combined (mixed method), a total of 23 species were recorded, which is 21% more than the observed species in either method (PAM or VES). Using bootstrap resampling, the 95% confidence interval was between 16 and 23 species (Fig. 3). The mixed-method richness estimate was significantly higher than the PAM estimate (W = 940992, p < 0.0005) and VES estimate (W = 939942, p < 0.0005). Similar to the accumulation curves of VES and PAM, sampling inadequacy was observed in the mixed-method SAC, where 0.53 additional species were detected with each new site. The extrapolated curve approached an asymptote at approximately three times the current sampling effort, suggesting that approximately 40 sites would be required to create a comprehensive inventory for the study area (Fig. 5).

In this study, both visual encounter surveys and passive acoustics yielded the same landscape-level species richness, with identical bootstrap-derived 95% confidence intervals. Similar results were observed in Andean forests, where equivalent amphibian species richness was observed across methods (Boullhesen et al. 2021). The PAM richness estimate was also similar to earlier visual surveys conducted in the Namdapha landscape. In 1984, Sarkar and Sanyal conducted an expedition and reported 14 amphibian species (Sarkar and Sanyal 1984). A more recent survey conducted in 2009 reported 22 amphibian species in NTR (Dinesh and Radhakrishnan 2019). Taxonomic updates in amphibian systematics over the years have led to revisions in the names of several taxa. Considering these changes, the PAM method in this study detected the following species that were not documented in earlier surveys: Microhyla eos, Hylarana leptoglossa, Nidirana noadihing, Gracixalus patkaiensis, and Raorchestes nasuta. Based on these findings, passive acoustic monitoring appears to be a promising alternative to VES for assessing and monitoring anurans, particularly in the vastly unexplored and dense rainforests of Northeast India. However, developing a comprehensive inventory for this region requires examining the species composition detected by each method.

The results showed that 15 species were common to both methods, while four unique species were detected by PAM and VES, respectively (Table 1). Analysis by ecological traits revealed that PAM was effective in detecting vocal amphibians that were either cryptic or occupied concealed niches. VES failed to detect Nasutixalus jerdonii and Zhangixalus smaragdinus, likely because they are arboreal and, hence, outside the observer’s field of view. Barnes and Quinn (2023) detected arboreal anurans from ARUs alone, highlighting the distinct advantage of PAM over VES in detecting tree-dwelling frogs (Barnes and Quinn 2023). Visually spotting species in the high canopy is challenging, and locating those hiding in subterranean environments is even more difficult. The ARUs detected a call type characterized by a low dominant frequency, which was attributed to a putative species of the genus Limnonectes based on its acoustic features. Boruah et al. (2026) formally described it as a new species (Limnonectes motijheel) and documented its calling behavior from mud nests beneath the leaf litter (Boruah et al. 2026). Similarly, PAM recordings in this study identified another previously undocumented call type. Based on its spectral and temporal features, it was inferred to be a cryptic species within the genus Raorchestes. This inference was subsequently supported by Boruah et al. (2025), who formally described the taxon as a new species (Raorchestes nasuta) (Boruah et al. 2025). With respect to call frequencies of the detected species, the highest species richness was observed in the mid-frequency range (2–4.5 kHz) across both methods. The most parsimonious explanation for this pattern is that mid-frequency callers may simply be more common in the assemblage. Only after accounting for this possibility should additional mechanisms, such as low-frequency masking or high-frequency attenuation, be considered. Although call masking is well documented in the literature (Wiley and Richards 1978; Buxton et al. 2017), its role cannot be confirmed in this study without targeted measurements of the acoustic environment. Overall, PAM outperformed VES by detecting calls from species that were otherwise difficult to detect visually, including arboreal, fossorial, and cryptic species. These detections also suggest that the landscape harbors complex environments with vertically structured amphibian habitats (Oliveira and Scheffers 2019).

Four amphibian species—Amolops adicola, Duttaphrynus melanostictus, Odorrana livida, and Limnonectes sp.—were uniquely detected by VES. Many Odorrana species produce complex ultrasonic calls (Shen et al. 2011; Arch et al. 2012). It is likely that Odorrana livida was not detected through PAM because its calls were outside the detection frequency range of the ARUs. Amolops adicola occurs in riverine habitats, with peak calling and breeding activity documented in October (Patel et al. 2021). Since this study was undertaken between May and June, it is likely that A. adicola was not vocalizing during this period for it to be detected by PAM. A greater number of species were detected by VES compared to PAM in stream habitats. This may be due to the high ambient noise around torrential streams, which is known to reduce acoustic detection rates (Towsey et al. 2014; Buxton et al. 2017). The Limnonectes sp. detected in VES was commonly found in road puddles and was not observed vocalizing during visual surveys. Similar to A. adicola, it is possible that this species breeds during a different season. The exclusive detection of D. melanostictus by VES may be attributed to its terrestrial and conspicuous nature, along with its low-frequency and low-intensity calls that may be masked by ambient noise, limiting its detectability by PAM. Echoing other similar findings, this study concludes that VES performs well in detecting amphibians that are either non-vocal, have low-intensity or infrequent calls (Dodd 2010; Guzy et al. 2014), or occur in habitats with high ambient noise, such as streams.

Noting the complementary nature of VES to acoustic methods, this study compared the performance of the mixed method (PAM+VES) with its counterparts. The mixed method yielded significantly higher amphibian species richness estimates for the studied landscape. These findings support combining visual and passive acoustic surveys for the comprehensive assessment of amphibian communities (Boullhesen et al. 2021; Hoefer et al. 2023).

The observed species richness in all three methods (PAM, VES, and PAM+VES) was equal to the upper limit of the bootstrap-derived confidence interval. This can be due to the presence of rare species—those detected at only one or a few sites. Nearly 50% of the total species detected in the mixed method were observed at a maximum of two sites. The presence of a high number of rare species in communities reflects habitat complexity and environmental heterogeneity, highlighting their conservation importance (Magurran and Henderson 2011). However, the percentage of rare species and the discrepancy in observed richness (vs. CI) could also be due to insufficient sampling. Limited or uneven sampling effort across the study area can result in a downward bias of bootstrap-derived richness estimates. In all three methods, non-asymptotic species accumulation rarefaction curves were observed (Fig. 5), reflecting the incomplete detection of community richness (Gotelli and Colwell 2001).

This study was designed to fulfill the stated objectives; however, it was constrained by logistical and financial limitations that influenced the extent of sampling effort. Although the full potential of PAM is typically realized through long-term deployments (Melo et al. 2021; Kunz et al. 2024), ARU deployments were limited to a single night per site due to resource constraints. Although the sampling effort was standardized to 6 h per site, this duration may be insufficient for detecting anurans that vocalize infrequently or exhibit temporal variation in calling behavior. Sampling was further limited to the onset of the wet season (May–June 2023) to minimize risks associated with landslides and flooding, thereby excluding the full seasonal variation. Spatial coverage was similarly limited, as sampling locations were based on previously surveyed accessible locations and did not encompass the full altitudinal range of the study area. Consequently, both temporal and spatial sampling efforts were restricted. Despite these limitations, PAM detected a diverse assemblage, including acoustically distinct taxa not recorded in earlier surveys, underscoring its sensitivity even under limited sampling effort.

Based on the extrapolated species accumulation curve, anuran species richness in the Namdapha landscape can be effectively estimated using passive acoustic devices by sampling across a minimum of 40 sites spanning different elevations and habitats. Future PAM studies should deploy ARUs for longer durations per site or incorporate temporally stratified recording schedules to test their full potential. Wherever possible, PAM should be complemented with VES, or sampling should be undertaken across multiple seasons of a year.