Introduction

Light pollution is increasing globally every year, and its sources are very diverse (Kyba et al. 2017). Streetlamps are considered to be the main source of artificial light at night (hereafter: ALAN), but researchers revealed that light emission from industrial, commercial, and public services can be responsible for more than half of the light output while covering less than one-third of the area (Kuechly et al. 2012). There is no scientific evidence on the emission of light pollution from greenhouses, but night sky activists emphasize how greenhouse lighting markedly intensifies light pollution (International Dark-Sky Association 2021). Greenhouses are often located outside cities, thus disturbing natural light conditions in a non-urban environment. ALAN has a significant impact on wildlife and one of the affected groups of animals are birds, whereby there are many studies confirming the disturbing effect of light pollution on annual and daily life cycles (Kempenaers et al. 2010; Da Silva et al. 2014; Dominoni and Partecke 2015). There is a lot of research on how ALAN affects birds in cities, but the impact on non-urban birds has been not well studied (Da Silva et al. 2016).

Changes in diurnal light intensities are one of the most important factors enabling movement and interactions in birds which leads to an associated increase in vocalizations (Davidson and Menaker 2003). Dim light stimulates vocalizations by causing neurological changes in the brain (Miller 2006), though ALAN is one of the ecological factors promoting nocturnal vocalization (La 2012), thus disrupting synchronization achieved by light-controlled internal clocks. Naturally early-singing species seem especially affected by ALAN because of bigger eye size they have a better ability to detect dim crepuscular light and start singing at lower light intensity than later less light-sensitive singers (Thomas et al. 2002; La 2012). Light at nighttime can affect not only singing behavior but also increased vocal activity in nocturnal migrating birds (Watson et al. 2016).

The aim of our study was to answer the question: how does greenhouse light pollution affect forest birds in a non-urban environment? We hypothesize that these artificial light sources influence the timing of birds vocal activity (singing and calling), but the effect should be more pronounced and observed for early singers.

Methods

Fieldwork

The study site was a 19 ha forest area (central point: 51.015621, 17.163041) near a 48 ha greenhouse complex in southwest Poland, 15 km from the city of Wroclaw. Greenhouses emit non-regular strong light pollution at variable hours between 4 a.m. and 11 p.m. from October to April. Before midnight, the lights at our site were turned off and the greenhouses were dark for most of the night and they were not turned on at all on some days. To compare results from the light-polluted site, we found a 25-ha dark control area with very similar tree species composition, located 16 km northwest from the center of Wroclaw (central point: 51.171346, 16.824681). Vocal activity of birds was recorded constantly from February 18 to March 1 and March 29 to April 4 in 2021 by a single Song Meter SM2 + Wildlife Acoustics recorder (Wildlife Acoustics, Inc., Maynard, USA) at each location. There was one recording at each site (ALAN and dark control). Devices were located ca. 20 m from the edge of the forest, on trees 4 m above the ground. Recorders were programmed (stereo, sampling rate 22 050 per second) to record dawn vocatizations of birds between 4 and 9 a.m. before and after sunrise and dusk chorus between 5 and 10 p.m. before and after sunset. The length of exposure to ALAN decreased as the season progressed from Winter to Spring—at later dates, there is an earlier sunrise and later sunset, and the greenhouse lighting time was not extended. In our recording period as the breeding season progressed, time of sunrise and sunset changed, respectively, 41 min earlier and 2 h 16 min later (according to the daylight saving time).

Data collection

Recordings were analyzed using Audacity 3.0.2 and RavenPro 1.6.1 (The Cornell Lab of Ornithology, Ithaca, USA). The time of each bird call before sunrise and after sunset was recorded to obtain when birds start/end their vocalization and how intensive it was (total number of calls/songs). After sunrise and before sunset (daytime light), only first calls and songs repeated at least three times during 5 min were noted to collect voices of all identifiable bird species and to record the time of the end of the dusk chorus. When species vocalized before dawn and after dusk, we did not include its daytime calls and songs in the analyzes. Data for dusk and dawn vocalization were collected separately each day. We categorized bird vocalizations as either song or call and record start/end time separately (Fig. 1). The schedule of emitting artificial light from the greenhouse was obtained using ALPS-WRO sky pictures from the Astronomical Institute, University of Wroclaw, available online (All-Sky Light Pollution Survey 2021). The ALPS-WRO camera takes sky pictures every 3 min. Light emitted from the greenhouses is strong enough that it is perfectly visible in the photographs on the southeastern site (Fig. 2).

Fig. 1
figure 1

Acoustic types used in the study. Spectrograms show differences between various types of Blackbird calls (A) and song (B) and Robin call (C) and song (D)

Full size image
Fig. 2
figure 2

Pictures of night sky taken by ALPS-WRO camera before (A) and after (B) turning the greenhouse lights on. Even in a cloudless night sky (B) in the bottom left corner skyglow caused by reflection of greenhouses light is visible. In cloudy weather, the effect of skyglow is enhanced

Full size image

Statistical analyses

We subset the data into three groups: for naturally early-singing birds—Blackbird (n = 478) and Robin (n = 53) and for naturally late-singing birds (other birds; n = 498). Our dependent variable was vocalization time, and it was calculated differently during morning and evening. For the morning, we computed the difference between sunrise time and time of vocalization. During the evening, we calculated the difference between time of vocalization and sunset time. Both variables were merged into one variable with day period as a categorical variable (morning or evening). Thus, positive values indicate vocalization before sunrise or after sunset (vocalization time in twilight or nightfall) and negative values indicate times after sunrise or before sunset (i.e., in daylight). Because of negative values, it was not possible to analyze data by a Poisson model, so we added 327 (the lowest value record) minutes to all records to obtain only positive values. For the Blackbird and Robin datasets, we used generalized linear models. For the other birds dataset, we used a general mixed model with species as a random factor (random intercept). For all three analyses, we used a negative binomial error distribution because we found overdispersion in the Poisson model. As independent variables, we used the factors: day period (morning, evening), season (February [18 February–1 March], March/April [28 March–4 April]), vocalization type (song, call), site (ALAN: Siechnice, control: Żurawiniec). For each model, we included interactions of site × day period × season and site × vocalization type × day period. We used such interactions because we wanted to test whether there was a difference between the control and ALAN sites in different months during morning or evening and between control and ALAN sites during morning or evening but depending on the vocalization type. For each model, we started from the full model with this structure. We tested the significance using the likelihood ratio test (LRT) which compares the full model to a reduced model where the target variable has been dropped according to AIC (Akaike information criterion); computation was performed in R with the function drop1(). We excluded all nonsignificant terms. If a 3-way interaction was not significant, we excluded it from the model and then checked 2-way interactions (site × vocalization type, site × day period). Post hoc comparisons (between sites for each day period in different months) were done using Sidak’s tests.

All analysis was performed in R software version 4.1.1 (R Core Team 2021) with MASS (Venables and Ripley 2002).

Results

For the first Blackbird analysis, the final model included the significant term: site × day period × season (df = 1, LRT = 7.659, p = 0.006). All model parameters are shown in Table S1 in supplementary materials. Post hoc comparison indicated that blackbirds began their dawn chorus earlier in the ALAN site compared to the control site only in February (p < 0.001; Fig. 3A). For other comparisons between sites in different day period and season groups, there were no significant differences (all p > 0.513). The interactions site × vocalization type × day period and site × vocalization type and the main effect of vocalization type were not significant (all p > 0.05).

Fig. 3
figure 3

Prediction of vocalization time (shown on the y-axis as day nightfall: positive values indicate vocalization timing in twilight [before sunrise/after sunset] or almost in night and the negative values indicate vocalization timing during daytime [after sunrise/before sunset). Panel A—model for Blackbird Turdus merula, panel B—model for European Robin Erithacus rubecula, panel C—model for other birds: naturally late singers. Panels show different interaction terms (A—site [ALAN: Siechnice vs control: Żurawiniec] × day period [morning vs evening] × season [February vs March/April]; B—site × day period × season; C—site × vocalization type [call vs song] with season as an additional effect, D—site × day period). The points represent raw data. Whiskers indicate 95% confidence intervals of prediction. The post hoc comparison (Sidak tests) in A–B was made between sites for each day period in different seasons and in panel C between sites in vocalization type groups. Asterisk indicates significant differences (***p < 0.001, *p < 0.05)

Full size image

For the second analysis with Robin data, the final model included the significant term: site × day period × season (df = 1, LRT = 6.891, p = 0.009). All model parameters are shown in Table S2 in supplementary materials. Post hoc comparison shows that Robin began their dawn chorus earlier in the ALAN site compared to the control site in February (p < 0.001; Fig. 3B). A similar post hoc comparison in March/April indicated a significant difference and the same relationship between sites (p = 0.049). For the other two comparisons between sites in the evening in both seasons, there were no significant differences in vocalization timing (both p > 0.528). For this analysis, we found that the main effect of vocalization type was significant (df = 1, LRT = 4.486, p = 0.034): timing of calls was more in the twilight/night time (mean = 23 min, 95% CI = 8–37; positive values indicate vocalization timing in twilight [before sunrise/after sunset] or almost in night) than songs (mean = 2 min, 95% CI =  − 5 to 10). The interactions site × vocalization type × day period and site × vocalization type were not significant (both p > 0.05).

For the third analysis with the other, later singers dataset the final model included the significant term: site × vocalization type (df = 1, LRT = 5.752, p = 0.017). All model parameters are shown in Table S3 in supplementary materials. Post hoc comparison indicates for song there was no significant difference in vocalization time between sites (p = 0.546). Post hoc showed also that the call vocalization timing was more intense closer to twilight (earlier in the morning and later in the evening) in the ALAN site than the control site (p < 0.001; Fig. 3C). However, day period was also significant as an additional effect (df = 1, LRT = 6.166, p = 0.013); during the evening, the vocalization time was performed more in daylight (mean = − 42 min, 95% CI = − 53 to − 30; the negative values indicate vocalization timing during daytime) than in the morning (mean = − 29 min, 95% CI = − 39 to − 20). We found that the main effect of season (df = 1, LRT = 9.176, p = 0.002) was significant: In February, the vocalization time was performed more in daylight (mean = − 43 min, 95% CI = − 53 to − 34) than in March/April (mean = − 28 min, 95% CI = − 39 to − 16). The interactions: site × vocalization type × day period, site × day period × season, and site × day period were not significant (all p > 0.05). The nonsignificant interaction site × day period is presented in Fig. 3D and shows that birds vocalizing time was similar in both sites in both periods of the day.

Discussion

The impact of all-night streetlamp light pollution on wildlife has been already well studied (Da Silva et al. 2014), but to date other sources of light pollution have rarely been examined. To the best of our knowledge, this is the first study on ALAN that shows how a short time (4–8 h per day) intensive light pollution emitted from greenhouses affects birds in a non-urban environment.

In our research, we have revealed that light pollution emitted from greenhouses affect daily vocal behavior not only in naturally early singers but also in late-singing bird species. In the ALAN site, Blackbirds began their dawn chorus earlier compared to the control site only in February (p < 0.001; Fig. 3A) and as the breeding season progress (March/April) effect decreased and birds stopped vocal respond to ALAN. Robins began their dawn chorus significantly earlier in the ALAN site compared to the control in the entire recording period (February: p < 0.001, March/April: p = 0.049; Fig. 3B). In both naturally early-singing species, there was no ALAN effect on dusk vocalization. Analyses of vocal activity in later singers (other bird species) revealed that birds vocalizing time was similar in both sites. There was no ALAN effect on dusk and dawn chorus in later singers (Fig. 3D). Due to different functions of vocalization types, we decided to analyze calls and song separately.

Singing is used to attract mates and defend territories (Kroodsma and Byers 1991). Calls aid individual recognition and communication with conspecifics and act as alarm signals (Marker 2004). While we found no effect of light pollution on singing time between sites (p = 0.546), there was ALAN effect on calling timing: Birds developed calls later in the evening and earlier in the morning than in the control site (p < 0.001; Fig. 3C).

Prolonged activity in birds exposed to ALAN can result in increased calling activity (Dominoni and Partecke 2015), extended foraging time (Da Silva et al. 2017), and overall higher energy expenditure (Ulgezen et al. 2019). As many studies have shown before, the impact of ALAN is stronger for the naturally early-singing bird species, such as the Blackbird and European robin than others (Kempenaers et al. 2010; Da Silva et al. 2014). In our research, we have confirmed these effects in the timing of dawn vocalization of both mentioned species. For Blackbird, the effect was only in the early season, but for Robin in the entire recording period. For dusk vocalization, there was no significant difference. This may be a compromise as birds cannot sing late at dusk and then start early at dawn. It would be too exhausting, besides Blackbirds naturally show the earliest dawn singing during peak female fertility (Da Silva et al. 2016). The effect of earlier vocalization was in the early season of the breeding season, probably because birds are the most sensitive to light variation as a result of photoperiodic induced gonadal growth and the peak of steroid hormone levels (Dominoni and Partecke, 2015). With the progress of the season, the sensitivity to light decreased, except for dawn vocalization of Robin. The strongest vocal response to ALAN of this species in the entire recording session had been already confirmed in other studies (Da Silva et al. 2014). This may suggest the better behavioral flexibility manifested by a greater vocal response even in conditions of substantially decreasing light pollution, compared to other species, e.g., Blackbird (Da Silva et al. 2016).

As we have assumed, short-term light pollution can affect the vocal behavior of other bird species, but only in call activity. Prolonged vocalization was observed when the lights were on throughout the day, as well as in the morning and in the evening. This suggests that ALAN can significantly disrupt the perception of natural day length, demonstrated by longer diurnal activity (Dominoni and Partecke 2015). Although other studies have confirmed that species such as Great Tit Parus major and Blue Tit Cyanistes caeruleus (dominant species on our research area) may respond with earlier singing to ALAN in non-urban areas (Da Silva et al. 2016), we did not confirmed this result with greenhouse lighting.

In our experiment, there would be possible to test whether the effects of ALAN from greenhouses cause immediate effects to vocalization timing which are reversible, or whether the effects persist after ALAN is removed. Unfortunately, the data collected are too scarce (in our research there were only three dark nights) to turn this experiment into a before-after-control-impact (BACI) design. In the future research on impact of light pollution emitted from greenhouses, it is worth considering BACI experiment.

To sum up, we showed that even non-regular, short-time intensive light pollution from greenhouses can affect vocal activity, not only of natural early singers, but also other bird species. However, there was only one study site per treatment, so these results cannot be generalized. More studies are needed with more replicates to investigate the cost and consequences of living near big sources of ALAN, such as greenhouses, for birds in a non-urban area, both during the breeding season and during migration.