Total solar eclipse triggers dawn behavior in birds: insights from acoustic recordings and community science

Abstract

On April 8th 2024, a total solar eclipse disrupted light-dark cycles for North American birds during the lead-up to spring reproduction. Compiling over 10,000 community observations and AI analyses of nearly 100,000 vocalizations, we found that bird behavior was significantly affected by these few minutes of unexpected afternoon darkness. More than half of wild bird species changed their biological rhythms, with many producing a dawn chorus in the aftermath of the eclipse. This natural experiment underscores the power of light in structuring animal behavior: even when ‘night’ lasts for just four minutes, robust behavioral changes ensue.

Humans have been captivated by total solar eclipses for millennia, and the 2024 “Great American Eclipse” on April 8 was no exception. April in North America is an interesting time for birds, many of which are preparing for their once-in-a-lifetime chance at reproduction. As tens of millions of people looked to the sky for their own once-in-a-lifetime experience, many wondered how wild birds would respond. After all, birds use light to time important behaviors on a seasonal and daily basis (1, 2). Since a total solar eclipse occurs in the same location once every three or four centuries (3), most free-living birds, like most people, have never seen day turn quickly to night, only to return again minutes later. Leveraging this natural experiment, we used autonomous recording units, machine learning, and community observations collected from Mexico to Canada to quantify how wild birds responded to this unexpected, and sudden, change in light.

Prior research lends some insight into this question, mainly via localized, anecdotal, or species-specific accounts (4, 5). For example, during a 1932 eclipse, New England residents described ‘subdued’ domestic poultry, cattle heading towards their barn, and frogs beginning their dusk chorus (6). In an observational study of zoo animals during the August 2017 eclipse, diurnal birds returned to evening roosts, whereas nocturnal species increased activity as totality approached (7). Also in 2017, this 2.5-minute eclipse changed the vocalization rate for four bird species in Nebraska, and a soundscape analysis showed an increase in species richness during and after totality (8). Big data approaches are also using solar eclipses as natural experiments on the power of light. For instance, Nilsson and colleagues used doppler radar to monitor biotic airspace during the 2017 eclipse, noting decreased activity before totality and more idiosyncratic bursts of activity during totality (9). Concurrently with our study of the 2024 eclipse, Mann and colleagues sampled bird vocal activity at over 300 locations (most in partial eclipse), finding that vocalizations were only affected in areas with at least 99% solar obscuration (10). Together, these findings emphasize that totality alters bird behavior, underscoring the need for new analyses of non-vocal behavior and species-specific effects.

Community science projects may resolve these uncertainties because they foster observations at a scale that is not otherwise possible. Such publicly collected data, also referred to as ‘citizen science’ or ‘participatory science’ (11), are accelerating discoveries in many fields, (12–16), including ongoing projects on wild birds (17). For the April 8, 2024 solar eclipse, totality lasted for two to four minutes, with the moon’s shadow racing ~2500 km in a couple of hours (Fig. 1). To capitalize on the intense human interest surrounding this brief event, we created a smartphone app called SolarBird, designed for users with zero birdwatching expertise. We publicized the app through local/national media, schools, and birding groups, gaining almost 11,000 observations. Users were asked to find a bird and observe it for 30 seconds, clicking up to 10 boxes to document its behavior (e.g. singing, flying, eating; Fig. S1). We analyzed these data in relation to totality, and the app calculated the sun’s obscuration via GPS and time stamps (Materials and Methods).

Fig. 1. Location of data collection.

Circles show unique SolarBird app observations submitted on April 8, 2024. Observations in the path of totality (100% obscuration; black) are the focus of our core analyses. Results from the path of partiality (5-99.9% obscuration; gray) are reported in Fig. S2, Table S2. Blue triangles (inset around Bloomington, Indiana, USA) denote locations for autonomous recording units.

To contextualize the app observations, we separately audio-recorded avian vocalizations before, during, and after totality on a localized scale. We were especially interested in sound because many birds use vocalizations to establish territories and attract mates (18, 19), and the timing of the April 8, 2024 solar eclipse overlapped with the annual peak of these behaviors during the lead-up to breeding in the lower Midwest of the United States. Avian vocal output often peaks daily around dawn and dusk (18), when light levels change most dramatically, particularly near the spring equinox in temperate regions (i.e., the time and location of this eclipse). Using autonomous recording units, we quantified all birds vocalizing within microphone range (20) of 14 locations in southern Indiana (Fig. 1, inset). Autonomous recording units have previously facilitated key insights on vocal behavior and biodiversity (21), and, coupled with machine learning (22), scientists can quantify vocalizations by species with limited human oversight (23). In particular, the artificial neural network BirdNET can identify over 6,000 bird species (23, 24). Though this technology is relatively new (25), it is already mapping avian diversity with fewer person-hours than traditional point counts (10, 26–29), thereby accelerating scientific discovery at previously infeasible scales.

Based on prior evidence from solar eclipses (4–9) and the observation that light is often used to entrain biological rhythms (1, 2, 30), we hypothesized that the total solar eclipse would disrupt normal bird behavior. Specifically, we predicted that birds would exhibit dusk-like behaviors during totality, followed by re-initiation of morning behaviors, like the dawn chorus, to coincide with the pseudo-dawn after totality.

Results and Discussion

Observations by the public show marked effects on bird behavior

We filtered SolarBird data to focus on 6,951 observations submitted by 1,174 distinct users in the path of totality on April 8 (Fig. 1 black circles, Data S1, Materials and Methods). Via user GPS and time stamps, we categorized observations as “before”, “during”, or “after” totality, the point when the sun was 100% obscured. We used binomial general linear models to analyze bird behaviors during each period. Vocalizations (songs or calls) were the most common observation, with users reporting significantly higher rates during totality than in periods before or after (Fig. 2a, Table S1). This pattern is consistent with the burst of vocalizations that some species exhibit around sunset. In contrast, observations of birds flying, remaining stationary, or other behaviors were documented at significantly lower rates during totality (Fig. 2bcd, Table S1). These totality-associated increases in vocalization and decreases in other behaviors may stem from observer bias; after all, light levels during totality were low (11.3 ± 1.1 lux, mean ± SD, Fig. S4), and observers may have reported what they heard while viewing the sun’s corona. However, vocalizations and flying were both more commonly reported after vs. before totality (Fig. 2ab, Table S1); this shift cannot be explained by differences in visibility. Instead, these results suggest that both behaviors increased after totality, consistent with a burst of activity after sunrise. Overall, these patterns were not found outside the path of totality (Fig. S2, Table S2), where partial obscuration ranged from 5.8-99.9% (Fig. S3).

Fig. 2. Bird behaviors documented by SolarBird app users in the path of totality.

Colored bars denote observations submitted on April 8 before totality (“Before”, red), during totality (“During”, black), and after totality (“After”, yellow). Reports of vocalizations and flying (a, b) were significantly different between all three periods, while reports of stationary and other (c, d) were significantly lower only during totality. *p < 0.05, **p < 0.01, ***p < 0.001. Exact p-values in Table S1. Results outside the path of totality are reported in Fig. S2, Table S2.

AI shows that some species are especially sensitive to light

Our own recordings also documented marked changes in bird vocalizations. Across 14 locations near Bloomington, Indiana (39.16°N, 86.53°W), we detected 52 species on the afternoon of April 8, 2024 (threshold for inclusion: high confidence and >10 unique detections across at least 5 recorders; Materials and Methods, Data S2). Totality began at approximately 15:05 Eastern Time (see Materials and Methods) and lasted for about four minutes, and we focused on the 60 minutes before and after totality. After processing these 124 minutes with BirdNET (Materials and Methods), we detected 15,606 vocalizations from these 52 species (Fig. 3, Data S3). This total was 20-50% more than what we detected during the same 124 afternoon minutes on the day before (10,322) or after (13,008). For a subset of species, we validated these AI detections with an expert birder (ELM), who scored excerpts by ear using the event-recording software JWatcher v1.0. Human-scored counts and AI-derived counts from these same time periods were significantly correlated (GLM: t(18)=3.588, p=0.002, d²=0.419; Fig. S7b). BirdNET detected fewer vocalizations than the trained and consistent human (Fig. S7), but BirdNET was a proportional reflection of human-scored vocalizations.

Fig. 3. Overview of vocalizations and analytical windows surrounding totality ± 1 hour.

Total vocalizations were summed across 52 species for the eclipse afternoon (green) versus equivalent times on three non-eclipse ‘control’ afternoons, with the dotted line showing the average and the gray ribbon showing the minimum and maximum on non-eclipse afternoons. Note, species-level data are the source for our formal analyses (Fig. 4), which focused four scoring windows: 12 min just before totality (“Before”, red); 4 min during totality (“Totality”, black), 12 min just after totality (“After”, orange); and 12 min slightly later when a delayed dawn chorus might be expected (“Later”, yellow).

Responses to the eclipse were not uniform among species. Using a Bayesian approach trained on the afternoons of April 6, 7, and 9 (Fig. S5, Data S3), we found that 29 of 52 focal species were affected in at least one time period (Fig. 4). In the 12 minutes before the eclipse (hereafter “before”), when light levels were especially different from a normal day (Materials and Methods), we documented significant effects on 11 species, most of which (10 of 11) vocalized at higher rates than usual for that time of day.

Fig. 4. Twenty-nine species responded to the eclipse.

Species are sorted and colored based on the period in which they were most affected, including 12 min just before totality (“Before”, red), 4 min during totality (“Totality”, black), 12 min just after totality (“After”, orange), and 12 min slightly later when a delayed dawn chorus might be expected (“Later”, yellow). See Fig. 3 and Materials and Methods for further detail on sampling periods. Each point represents an effect size (scaled per minute) with 95% CI, colored if significant. All 52 species, including those not significantly affected, are shown in Fig. S6.

During totality, 12 species were affected, with half decreasing and half increasing vocalization rate, relative to non-eclipse ‘control’ afternoons. American robins (Turdus migratorius) displayed the strongest increase during the eclipse (nearly 5x higher than a typical afternoon). Considering that robins are the most abundant in our recordings and also ubiquitous across the USA, it is possible that the robin’s vocal burst may partially account for the significant increase in vocalizations reported by app users.

The largest number of species (n=19) significantly adjusted their vocalization rate after totality ended, either in the 12 minutes immediately after totality (“after”) or the 12 minutes centered around 30 minutes after totality when a delayed dawn chorus might occur (“later”). Nearly all post-eclipse changes were significant increases relative to a non-eclipse afternoon, consistent with a pseudo-dawn chorus. Notably, barred owls (Strix varia) produced 4x as many calls on the eclipse afternoon compared to control afternoons, and American robins, which are notorious for their boisterous predawn chorus, vocalized at 6x their normal rate just after totality.

Dawn chorus behavior is repeatable and predicts eclipse behavior

Next, we asked whether dawn and dusk behavior predicted behavior immediately before or during totality (akin to dusk) as well as immediately after totality or ~30 minutes later (akin to dawn). Using the two most recent dusks (April 6 and 7) and dawns (April 7 and 8), we used BirdNET to score 2-hour periods centered on the transition between civil twilight and sunrise or sunset. Using all the same parameters and thresholds as before, this yielded 47,932 more vocalizations (Data S2, Data S3). Vocalization rates (per min per species) were consistent across consecutive dawns and consecutive dusks, with similar onset and offset times each day (Fig. S8). For instance, American robins and barred owls vocalize most when it is still dark out, tapering by morning civil twilight and increasing again near evening civil twilight (Fig. S5). Eastern towhees and European starlings exhibit a burst of vocalization in the morning around civil twilight and sunrise, respectively, but around sunset, only towhees increase their vocalization (Fig. S5). Still other species vocalize continuously or off-and-on throughout a typical day, with no crepuscular peak (e.g. house sparrow, northern cardinal; Fig. S5). These findings reinforce common knowledge among birders, that each species has a relatively stereotyped vocal response to naturally occurring changes in light.

We operationalized the dawn (or dusk) chorus as ‘present’ if the species’ per minute morning (or evening) peak showed a 100% increase relative to the afternoon (Fig. S9, Fig. S10) and used binomial regressions to assess whether dawn or dusk behavior predicted eclipse behavior (Materials and Methods). Because species with a robust chorus naturally display the greatest behavioral shift associated with everyday changes in light, we expected them to be overrepresented in those species affected by the eclipse. This prediction was not supported for dusk behavior, which was not predictive of eclipse behavior in the 12 minutes before (GLM: z(51)=1.158, p = 0.25) or four minutes during totality (GLM: z(51)=0.951, p = 0.34). However, species with a dawn chorus were more likely than expected by chance to exhibit a significant change in vocalization in the 12 minutes after totality (GLM: z(51)=2.023, p = 0.04) but not in the 12 minutes centered around 30 minutes after totality (GLM: z(51)=0.458, p = 0.65). In essence, even though the false night only lasted for four minutes, species that naturally produce a dawn chorus immediately began a new dawn chorus when the false night ended.

Implications

Changes in light surrounding dawn and dusk are thought to be the primary cue structuring circadian changes in behavior for most organisms (30). Here, we show that, when the light-dark cycle was altered by a total solar eclipse, entrained biological rhythms were disrupted for over half of the observed wild bird species. This natural experiment – which essentially switched off the sun for four minutes in the middle of an otherwise normal spring afternoon – is unique for several reasons. For one, light experiments often start from constant light or constant dark (30, 31) in laboratory environments that may not recapitulate natural behaviors or physiology (32). The total solar eclipse, however, offers a large-scale manipulation that is both naturalistic and decoupled from time of day. Together, our AI- and app-derived data both underscore that the return of light is an especially salient cue, even when that switch occurs at an unexpected time (half a day early) and after an exceptionally short ‘night’ (0.5% the length of the previous night). Eclipse-associated changes in light are also decoupled from other major environmental shifts that accompany ordinary mid-day changes in light (e.g. thunderclouds), though atmospheric pressure, temperature, and even plant photochemistry change surrounding a total solar eclipse (33) and may have played a role (but see (10)).

The reverse experiment – artificial light at night, or ALAN – affects everything from behavior to biodiversity, mediated via sometimes subtle changes in biological rhythms (34–36) and neurophysiology (31, 37, 38). With ALAN of just a few lux affecting ecosystems worldwide (31), we urgently need to understand why some organisms are more affected than others. Our results provide unique experimental evidence on these interspecific differences in light sensitivity (Fig. 4), including 23 species that were spread across different avian orders but were not significantly affected by the eclipse (Fig. S6). Though it is possible some species may have switched the nature of their vocalizations (e.g. from mating to alarm calls), their otherwise behavioral insensitivity implies sensory mechanisms that maintain biological rhythms despite unusual changes in light.

Our study also highlights the underused ability of AI and community science to enrich one another. For example, publicly collected data can sometimes suffer from sampling bias (12), technology barriers (15), or quality control (14), yet our simple smartphone app was designed for users without specialized background or equipment. With their repeated observations, we learned that eclipse-related changes in vocalizations were accompanied by changes in movement (Fig. 2), and elements of our localized AI findings apply at a continental scale. In turn, our recordings provide species-level results (Fig. 4) that expand the public’s observations (Fig. 2). Moving ahead through the Anthropocene, these technologies promise to lower the bar for active public engagement in science, while deepening our understanding of the natural world.

Data and materials availability:

All data are available in the main text and/or supplementary materials, with source code for SolarBird app development and scripts available on Zenodo and Github (39, 40).