Regional atmospheric conditions

We derived atmospheric conditions based primarily on weather data from observations at five National Weather Service Automated Surface Observing System (ASOS) stations occurring within 50 km of the centroid of microphone locations (Figure 1). ASOS reports wind speed and direction (recorded approximately 10 m above ground level), as well as precipitation amount, every minute, although the data are derived from accumulations over the previous 1 or 2 min; visibility, cloud cover, and cloud ceiling data were reported every 5 min. We calculated wind profit from wind direction and wind speed [39]; wind profit represents the distance a bird is drifted toward a specified target direction in a fixed time interval through only the effect of wind. Typically, the target direction is the migratory goal, but we specified due southeast (135^o) as the target direction to better capture those combinations of wind direction and speed that indicate recent cold front passage and are also more likely to induce a coastal flight path, and perhaps the coastal effect [18], in migrating birds. We calculated nightly averages for weather variables from evening to morning civil twilight, thus encompassing the period of active monitoring. We also calculated the proportion of hours during a given night with at least one ASOS station reporting precipitation. Additional details of weather data acquisition and manipulation are available from the authors.

Analysis

We used generalized additive models (GAMs; [40], [41]) to explore the association between regional atmospheric conditions on NFC detections and differences in NFC rates among sites. GAMs accommodate potential nonlinear changes in calling activity with predictor variables while allowing us to incorporate serial correlation [41]; we implemented them using the gamm function of the mgcv package [42] in R, version 2.15.2 [43]. We estimated GAMs separately for warblers and sparrows. For each group, we modeled the number of nightly NFC detections as a sum of explanatory variables using a GAM with the following formulation:where NFC ∼ NegativeBinomial(θ), β₀ is an intercept, and the remaining βs describe their associated explanatory variables; β_site and β_year are categorical variables and comprise seven and two parameters, respectively. We fit a single smooth term (f_{season, site}) to allow for potential nonlinear seasonal (i.e., day of year) effects using the default thin plate regression spline; we allowed this smooth to interact with the categorical site variable, resulting in the possible estimation of separate smooths for each site. We did not evaluate any additional interactions to avoid over-parameterizing the model relative to the data, nor did we expect interactions among atmospheric conditions to possess much ecological relevance. We estimated the θ parameter for the negative binomial distribution with a similarly-structured generalized linear model, although replaced the spline for seasonal effects with a third-order polynomial. A first-order autoregressive (AR-1) error structure reasonably accounted for serial correlation in residuals. We grouped the correlation structure within each site and year combination to expedite GAM estimation [41]. We centered and scaled by one standard deviation all continuous model input variables to improve estimation and facilitate the assessment of the relative importance of atmospheric conditions to NFC detections [44]; we did not modify the categorical site or year variables. We omitted cloud ceiling from consideration prior to analysis due to its high collinearity with cloud cover (i.e., variance inflation factor >10; r = −0.97, df = 124, P<0.0001). To avoid biased parameter estimates and standard errors when evaluating hypotheses, we did not eliminate any terms from the models [45]. Finally, we estimated the average seasonal discrepancy in NFC detections between mainland sites and island sites with two additional GAMs (i.e., warblers and sparrows) that dichotomized recording sites according to this geographic context.

We compared automated detections with manually quantified NFCs in a sample of 10-minute recordings to evaluate the effectiveness of our automated detector settings. We selected 12 recordings from each site, with two notable constraints: recordings contained ≥4 warbler or sparrow NFCs (median  = 7; maximum  = 45) and occurred on separate nights. We again ignored NFCs too weak to assign confidently to warblers or sparrows, or NFCs belonging to other bird families. Thus, we estimated detector efficiency under generally favorable recording conditions and compared efficiency among sites and across a restricted range of ambient noise conditions. We modeled the logit of the proportion of successfully detected NFCs in each recording (P_detect) as a function of two explanatory variables, and weighted by the number of manually detected NFCs, using a generalized linear model (glm function in base R) with the following formulation:where β₀ is the intercept and the remaining βs describe their associated explanatory variables. We sampled from all eight sites, so β_site comprises seven parameters. The noise variable represents the average power (dB) in the 6–11 kHz frequency range in each 10-minute recording.

Associations of NFC detectability with atmospheric conditions

Atmospheric and ambient conditions can influence the detection rates of NFCs directly by inducing a change in the rate at which migrants call, or indirectly by influencing the number of birds aloft or NFC detectability. Certain atmospheric conditions commonly are associated with increased numbers of birds aloft in north temperate areas (e.g., front passage, wind conditions, and precipitation; [1], [2], [4]) and increase the likelihood of migrant concentrations along the Atlantic Coast of North America [22], [28]. We formulated wind profit to reflect wind conditions favorable for migration in general (i.e., the northerly component that typically follows cold front passage and building high pressure) but also to favor a westerly component more likely to induce a flight path towards the coast and offshore displacement. While northeast and east winds may provide favorable tailwinds to migrating songbirds in our region [2], [4], we expected winds from these directions to diminish the concentrating influence of the Atlantic Coast on southbound migrant activity. Consequently, our formulation of wind profit ascribed reduced or negative values to these wind conditions. We suggest the strong association of warbler and sparrow NFC detections with wind profit supports the idea that NFC detections generally reflect the number of birds aloft. We note that in addition to wind profit, several alternatives exist for capturing the multivariate problem of wind assistance, each making different assumptions regarding the behavior of the organism of interest [46].

Unlike wind profit, the extent to which other atmospheric conditions influence NFC detections directly or indirectly is less clear and not likely to be mutually exclusive. Therefore, atmospheric conditions likely complicate the relationship between the number of vocal birds aloft and NFC detections. For example, precipitation is thought to suppress migration [1], [2], [4], which concurs with the negative association of regional precipitation with NFC detections we documented in this study. However, consistent precipitation also compromises the detectability of NFCs by increasing the background noise (see below). In contrast, light to moderate precipitation may decrease visibility without hindering migration and thus induce increased calling rates [35], [47] or lower flight altitudes [8], thereby increasing detections.

Cloud cover and visibility describe additional conditions under which the influence on NFC detections may be direct or indirect. NFC detections are known to be positively related to increasing cloud cover (or decreasing cloud ceiling, which correlated very strongly with cloud cover in this study) or decreasing visibility (reviewed in [35]; see also [47]). Farnsworth [35] suggested that increased calling rates by individuals under conditions of poor visibility may be adaptive for maintaining contact, avoiding collisions, and coordinating migratory behavior, particularly in inexperienced migrants [48]; this hypothesis implies a behavior-modifying influence on calling rates. Its potential pertinence to inexperienced migrants is particularly relevant to this study, as the vast majority of autumnal songbird migrants along the coast are young birds performing their first migration [49]–[53]. But indirect effects seem equally plausible; poor visibility, cloud cover, and low cloud ceiling may decrease flight altitudes relative to clearer nights, placing more migrants within NFC detection range. Although we documented the expected increase in NFC detections with increasing cloud cover (warblers and sparrows) and decreasing visibility (warblers only), we were unable to distinguish between changes in calling or flight behaviors.

We further expected certain atmospheric and biological conditions to increase background noise and decrease our ability to detect NFCs, and thus associate negatively to NFC detections. Wind and insects comprised the primary sources of noise in this study. Wind speed associated strongly and negatively with NFC detections. The strong correlation between wind speed and background noise measurements at each site (r = 0.59, P<0.001, df = 622; ‘within-site’ correlation sensu [57]) suggests wind noise importantly reduced the detection of NFCs. Nonetheless, higher wind speeds might also decrease the number of migrants aloft, particularly when opposing the direction of travel [4], [54], although high wind conditions may also result in lower flight altitudes [54]–[56] and thus possibly increase detectability. We suggest that the negative effect of wind speed reflected primarily an increase in background noise and an associated decrease in the detectability of NFCs more so than a decrease in warbler and sparrow abundance aloft [34]. Insect noise was the primary non-wind source of background noise above 6 kHz. Our manual detector evaluation revealed a clear reduction in NFC detection due to insect noise in some instances (A. D. Smith unpubl. data), but its irregular and ephemeral occurrence made it impractical to quantify on a nightly basis. Insect noise occurred less commonly as the season progressed. While high background noise levels universally impaired NFC detections, intermediate noise levels may have disproportionately discriminated against species that produce calls not contained completely in the frequency range of our detector or less powerful calls, or that fly at higher altitudes. For example, most warbler NFCs occur above 6 kHz, but a few common species produce flight calls that regularly drop below 6 kHz (Table 2). We suggest that these calls were more sensitive to noise levels given our detector settings. In contrast, the NFCs of all sparrow species in this study occur entirely above 6 kHz [30]. This perhaps explains the weak influence of non-wind ambient noise on warbler NFC detection and the apparent lack of such an influence for sparrows.

NFC detections and coastal context

We recorded warbler and sparrow NFCs in multiple coastal contexts along the Atlantic Coast of the northeastern United States, an important migratory corridor during autumn migration. NFC detections occurred episodically over the fall migration season, similar to migration intensity in the region (e.g., [9], [15], [17], [58]), presumably the result of most migrants coinciding movements with ephemerally favorable conditions [2], [4], [28], [39], [59]. Although offshore islands often offer excellent opportunities to observe high densities of migrants [20], [37], we expected that most migrants would move over land or near the coast, rather than offshore, and thus would detect more NFCs at mainland sites compared to offshore sites. Indeed, warbler and sparrow NFC detections were considerably higher at our two mainland sites. Radar studies suggest the difference in NFC detections between coastal contexts along the Atlantic Coast reflects migrant abundance rather than a difference in the calling behavior or flight altitudes of birds, as the bulk of migration intensity occurs along the coast and inland, not over ocean, excluding water crossings from Nova Scotia over the Gulf of Maine [15]–[17], [60]. Furthermore, if patterns of NFC detections reflected changes in calling behavior more so than abundance, we might reasonably expect the opposite pattern (i.e., birds displaced offshore increase calling rates). Flight altitude might play some role in the observed pattern, but the data are scarce and mixed and often complicated by radar peculiarities (see, e.g., [8], [21], [25]).

There exists a well-documented pattern in ‘on-the-ground’ migrant densities on Block Island: migrants occur in higher densities on the northern half of the island, where they prepare for reoriented flights to the mainland or subsequent migratory flights [20], [37]. Indeed, two migration banding operations on the island ([52]; USFWS unpubl. data) exploit the phenomenon, as have multiple previous studies [61]–[65]. We thus expected a similar latitudinal pattern in NFC detections among Block Island sites; however, we found little evidence for this pattern (or differences among sites in general), suggesting that concentrations of migrants on the northern half of Block Island result primarily from redistribution after landfall (e.g., [22], [24], [66], [67]). Indeed, regular observation of significant diurnal, northerly movements of migrants on offshore islands following nights of active southerly migration provide evidence of such a redistribution ([20], [37]; A. D. Smith pers. obs.).

Migration activity along the Atlantic Coast, as assessed by radar, generally peaks in the few hours following sunset and declines steadily thereafter [9], [33], [58], [60], [68]. Comparisons of NFC detections with radar are few but suggest that NFC detections follows a similar pattern ([34]; but see the New York data in [33]) or peak up to a few hours later, usually near or just after the middle of the night [33], [69]. Seasonal patterns of intra-night NFC detections in this study (Figure 4) support an apparent delay in peak NFC detections relative to expectations from previous radar work. The patterns also suggest that sparrows migrate, or at least call, slightly later in the night on average than warblers (Figure 4). The intra-night patterns of warbler and sparrow NFC detections were generally consistent for the larger coastal context (i.e., mainland vs. island) and among latitudinal contexts on Block Island, with the possible exception of sparrow NFC detections at southern Block Island sites (Figure 4D). For reasons that remain unclear, the reduced activity near sunrise and sunset at these sites suggests that fewer sparrows are landing and settling on southern Block Island. Finally, rather than comparing seasonal averages of intra-night activity, more detailed work will be necessary to evaluate the variability of the relationship between migrant density and concurrent NFC detections (e.g., [34]).

NFC species composition

Acoustic monitoring is relatively inexpensive compared to radar, the equipment can be automated, and it provides information not readily obtained from other methodologies, including species composition and phenology information for vocal species and the ability to detect secretive, rare, or species otherwise difficult to survey [70], [71]. However, inferring the relative abundances of calling species using patterns of NFCs is complicated because several common species do not regularly vocalize during migration (e.g., flycatchers, vireos, mimids; [30], [35], [36]) whereas other species regularly vocalize during migration and so may be over-represented in NFC recordings (e.g., Savannah Sparrows; [71]). For example, a comparison of NFC detections at four of our microphone locations with capture rates at active and close (<500 m) banding operation suggests that Savannah and Chipping Sparrows were likely over-represented in NFC recordings whereas Yellow-rumped Warblers were likely under-represented in NFC recordings. Clearly, inferring the relative abundances of calling species using patterns of NFCs requires more knowledge of calling rates among species [36] and species-specific differences in NFC detection. Despite these potential complications in interpreting NFC data and inferring migration intensity, the acoustic monitoring of NFCs provides a viable and complementary methodology for exploring the spatiotemporal patterns of songbird migration (e.g., [29], [71]; see also oldbird.org), as well as evaluating the atmospheric conditions that shape these patterns.