Wildfires are an integral part of many ecosystems, shaping and influencing biodiversity patterns across diverse landscapes. On a global scale, approximately 4% of the land surface burns annually; in North America, an average of 7.2 million acres of land burned each year over the past decade (NIFC 2024). The frequency, intensity, severity, and size of fires in North America is increasing, largely due to climate change and changes in land use (Dennison et al. 2014; Fauria and Johnson 2008; Schoennagel et al. 2017). While most fire ecology research has focused on the direct impacts of habitat loss and transformation, wildfires also have far-reaching consequences beyond their burn perimeters, including air and water pollution (Garcês and Pires 2023; Moritz et al. 2014).

Smoke from wildfires, particularly those originating in Canada, has increasingly blanketed the northeastern U.S., including New York State (Hung et al., 2020; Tanne 2023), raising concerns of widespread impacts on public and ecosystem health (Cascio 2018; Chen et al. 2021; Jaffe et al. 2020; Johnstone et al. 2016; Pereira et al. 2021). For example, in June and July of 2023, smoke from boreal wildfires led to sustained periods of degraded air quality across the northeastern U.S., resulting in some of the worst air pollution ever recorded in New York State (Boulanger et al. 2024; Jain et al. 2024; Yu et al. 2024). Daily mean concentrations of fine particulate matter (PM_2.5), a primary component of smoke pollution, exceeded 120 µg m³ across much of the state – the highest levels recorded in New York in more than 50 years (Wang et al. 2024; Figs. 1 and 2). Historically, air quality in the northeastern U.S. has improved due to regulatory efforts aimed at reducing anthropogenic emissions (Hung et al. 2020 b), but recent large-scale wildfires in Canada have interrupted this trend (Fig. 1). Given the increasing frequency and severity of wildfire smoke events – particularly in places where such disturbance was previously uncommon – evaluating the ecological consequences of wildfire smoke is of critical importance to conservation (Sanderfoot et al. 2021).

New York State provides a useful case study for understanding how wildlife may respond to episodic smoke exposure in regions that do not typically experience severe wildfire activity. Birds are valuable indicator species for ecosystem health and may be particularly vulnerable to wildfire smoke due to their highly efficient respiratory system (Brawn et al. 2001; Sanderfoot and Holloway 2017). Acute exposure to wildfire smoke can lead to respiratory distress (Chao et al. 2019), reduced lung function (Demling 2008), and myriad other adverse health effects in air-breathing animals, including birds (Sanderfoot et al. 2021). Though research is limited, emerging evidence suggests that smoke has wide-ranging impacts on avian health and behavior. A recent study linked chronic exposure to wildfire smoke to loss of body mass in wild birds, suggesting declines in body condition (Nihei et al. 2024; Wohlsein et al. 2016). Birds may also alter their activity, movement patterns, or habitat use during smoke events, potentially as a response to underlying physiological stress. For example, Greater White-fronted Geese (Anser albifrons) modified their migration routes and flight durations in response to dense smoke (Overton et al. 2022), and capture rates of passerine and near-passerine birds at a long-term banding station declined with acute smoke exposure (Nihei et al. 2024). Previous work has demonstrated that wildfire smoke can influence bird detectability during surveys, likely due to changes in bird behavior (Sanderfoot and Gardner 2021). However, most research on avian responses to smoke has focused on birds in western North America, where wildfire smoke events are relatively common, and few have leveraged large-scale biodiversity data repositories.

Smoke events are highly unpredictable in terms of timing, intensity, and duration, making it challenging to design structured field studies that accurately assess their transient impacts on birds. As such, citizen science – or public participation in scientific research – has emerged as a valuable tool to explore the effects of smoke pollution on bird populations in real time (Sanderfoot and Gardner 2021, Sanderfoot et al. 2022). eBird is a global participatory science program run by the Cornell Lab of Ornithology that supports opportunistic, semi-structured collection of data on bird observations by the public (Sullivan et al. 2014). Despite the challenge of observer variability and data quality issues, rigorous data filtering protocols and best-practice modeling approaches have been shown to yield reliable ecological inferences from eBird data (Johnston et al. 2021; Walker and Taylor 2020). eBird data have been widely used to describe species distributions (Coxen et al. 2017; Dennis et al. 2017; Emmet et al. 2023; Fink et al. 2024; Fournier et al. 2017; Steen et al. 2019), support long-term population monitoring (Horns et al. 2018; Johnston et al. 2025; Neate-Clegg et al. 2020; Walker and Taylor 2020), and analyze bird movements (Fuentes et al. 2023; Martín et al. 2020). eBird data was previously used to examine how wildfire smoke impacts observations of birds in Washington State, a region that encompasses both fire-prone ecosystems and ecosystems with long fire-return intervals (Sanderfoot and Gardner 2017).

In this study, we used data from eBird to assess the effects of wildfire smoke on the probability of observing birds in New York State during the 2021–2023 breeding seasons. We hypothesized that elevated ambient PM_2.5 concentrations, a marker of wildfire smoke, would generally reduce bird detection probabilities, potentially due to changes in bird activity, movement patterns, or observer detection conditions. We expected that the relationship between PM_2.5 and detectability would vary among species, as species-specific responses to smoke are likely mediated by physiological, life history, and ecological traits (Cornelius et al. 2026), including habitat association, foraging strategy, and the sensory modalities through which birds are detected during surveys. Forest-dwelling insectivores are acoustically cryptic and detected primarily through vocalizations; conditions that reduce vocal activity may therefore disproportionately reduce their detectability (Lee et al. 2017; Sanderfoot et al. 2024). Aerial insectivores, which are frequently detected visually, may adjust their activity in response to changes in insect-prey availability associated with wildfire smoke (Liu et al. 2022), potentially influencing observer detection rates. Understanding how traits mediate responses to wildfire smoke and influence our observations of birds in smoky conditions is important, both for interpreting biodiversity monitoring data and anticipating which species – and monitoring programs – are most vulnerable to smoke disturbance.

Study area

New York State is home to a diverse range of ecosystems, including temperate forests, wetlands, grasslands, and alpine habitats (Hess et al. 2024). While wildfires are not as common in New York as in the western U.S., fire plays a key role in shaping specific ecosystems, such as the pine barrens of Long Island and the fire-dependent oak-hickory forests of the Hudson Valley (Kurczewski and Boyle 2000; Lorimer 1981). These fire-adapted habitats rely on periodic burning to maintain biodiversity and prevent the encroachment of woody vegetation. Wildfires in New York are relatively small, with approximately 2,100 acres burned annually, and do not typically result in large-scale smoke events. In contrast, long-distance smoke transport from large wildfires in the western U.S. and southern Canada can impact air quality in New York and other eastern U.S. states (Hanes et al. 2019; Shrestha et al. 2022). One of the primary concerns associated with wildfire smoke is inhalation exposure to fine particulate matter (i.e., airborne particles with a diameter of 2.5 micrometers or smaller; PM2.5), a pollutant with well-documented adverse effects on human health (O’Dell et al. 2020) and emerging evidence of risks to wildlife (Sanderfoot et al. 2021). In June and July 2023, smoke from boreal wildfires contributed to poor air quality conditions throughout New York State (Figs. 1 and 2), with daily mean PM_2.5 concentrations exceeding the World Health Organization’s (WHO) 24-hour standard of 15 µgm^–3 (WHO, 2021) on multiple occasions, at times exceeding 120 µgm^–3 – eight times the value deemed unsafe for humans (Fig. 2). These spikes far exceeded typical urban background levels and correspond to known periods of transboundary smoke transport from Canadian wildfires (Yu et al. 2024), strongly implicating wildfire smoke as the dominant source of particulate pollution during this period.

Bird observations

eBird supports opportunistic collection of semi-standardized data on bird observations submitted by volunteers in a checklist format. Checklists also include information on the detection process (e.g., protocol, duration, start time, distance traveled, etc.) (Sullivan et al., 2009). We focused our study on observations of birds included in eBird checklists (Fig. 3) submitted during the breeding season (May 1–August 31) in the state of New York for the years 2021–2023. We utilized the auk package (Strimas-Mackey et al. 2018) in R (R Core Team 2023) to filter bird observations to include only complete surveys with stationary, traveling, or area protocols, for which birding is the primary focus. Restricting the dataset to complete checklists allows species not reported on a checklist to be treated as true non-detections, a widely adopted approach for generating presence–absence data from eBird observations (Johnston et al. 2021). Stationary surveys require birders to remain within about 100 feet (30 m) of the starting point of their checklist, whereas traveling surveys are conducted along a route extending more than 100 feet (30 m) from the starting point. The area protocol is designed for specialized surveys that thoroughly search an area of known size. Among the three protocols, only traveling surveys provide distance information. We assigned a distance of zero kilometers to stationary surveys. Following Sanderfoot et al. (2021), we calculated the distance traveled for area surveys based on the size of the area surveyed. Since the surveyed area is assumed to be a square, the distance traveled can be estimated as approximately twice the length of one side of the square. Since the side length of a square is equal to the square root of the area, we calculate the distance traveled using this relationship.

We used the auk package to zero-fill all checklists, generating detection/non-detection data for each species (Johnston et al. 2021). Zero-filling converts eBird checklists into presence–absence observations by assigning non-detections for species that were not reported on complete checklists. Because eBird group checklists may contain identical observations submitted by multiple participants, the filtering process retains a single representative checklist for each group to avoid duplicate records. To ensure sufficient statistical power and model convergence, we filtered our dataset to include only species recorded as present on at least 600 checklists in the baseline year (2021), a threshold comparable to that used in previous analyses of smoke effects on bird detectability (Sanderfoot et al. 2021). This threshold was applied after filtering checklists by survey protocol, completeness, and geographic extent to ensure that the retained species had sufficient observations distributed across space and time within the study region. Eighty-four bird species met this criterion (see Supplementary Table 3 for the full species list). Because each eBird checklist represents a unique survey event associated with a specific location and observer, the retained detections reflect observations from many independent sampling locations across New York State.

Air quality data

Daily exposure to wildfire smoke is often assessed by monitoring PM_2.5, a major component and well-established marker of wildfire smoke (Burke et al. 2022; Jaffe et al. 2020). We focused our analysis on the breeding seasons (May–August) in 2021–2023. Although the most extreme episodic smoke occurred during spring and summer of 2023 (Fig. 2A), we retained observations from all three years to capture the full range of background and elevated PM_2.5 concentrations and to evaluate how day-to-day variation in particulate pollution relates to bird detectability. To estimate air pollution exposure for each eBird checklist, we utilized data from the Environmental Protection Agency (EPA) Air Quality System (AQS) (US EPA 2013). Specifically, we downloaded daily mean PM_2.5 concentrations for all ground-based air quality monitoring stations operating in New York during 2021–2023 from the Air Quality System website (US EPA 2016). Each checklist was linked to the daily mean PM_2.5 concentration recorded at the monitoring station nearest to the checklist location.

We further filtered the dataset to include only observations associated with an AQS parameter code of 88,101, which represents measurements of PM2.5 used for regulatory monitoring under the National Ambient Air Quality Standards (NAAQS). Restricting the dataset to this parameter ensures that PM2.5 values were obtained from standardized Federal Reference Method (FRM) or Federal Equivalent Method (FEM) instruments used in regulatory monitoring networks. To ensure consistency with the spatial resolution of our weather data (25 km) and to minimize spatial mismatch between monitoring stations and eBird checklist locations, we excluded checklists located more than 25 km from the nearest air quality monitoring station. This approach ensured that air quality estimates were spatially comparable with other environmental covariates used in the analysis. Local PM2.5 concentrations reflect both background air pollution and short-term episodic increases in particulate matter associated with wildfire smoke.

Land cover data

We included land cover as a categorical predictor to account for broad habitat differences among eBird checklist locations across New York State that could influence detectability of birds. Land cover data were obtained from the CONUS National Land Cover Database (NLCD) (MRLC 2025) which provides land cover classifications at a spatial resolution of 30 m. The original NLCD classes were aggregated into eight categories representing major habitat types relevant to birds in the region: Barren, Cultivated, Developed, Forest, Herbaceous, Shrubland, Water, Wetlands, and Ice/Snow. All developed classes (e.g., “medium-developed,” “highly-developed,” and “open-space developed”) were combined into a single “Human-Modified” category.

Each eBird checklist was assigned a land cover class based on the NLCD pixel intersecting the geographic coordinates associated with the checklist location. This approach provides a consistent point-based habitat classification across all checklists and is commonly used in analyses of eBird data when habitat is included as a categorical covariate. Our goal was not to quantify detailed habitat composition surrounding each survey, but rather to control for broad habitat differences among checklist locations that may influence detectability of birds. Any checklists associated with unclassified pixels were removed from the dataset. Checklists labeled as “Ice/Snow” were also excluded due to the very small number of observations in this category.

Weather data

To account for the potential influence of weather conditions on bird detectability, we included air temperature and relative humidity as predictor variables. Weather data were obtained from the ERA5 reanalysis dataset provided by the Copernicus Climate Change Service, which has a spatial resolution of 0.25° (approximately 25 km) (Copernicus 2025). We used hourly estimates of temperature and relative humidity associated with each checklist observation. Relative humidity was included as an indicator of atmospheric moisture conditions, which may influence bird activity and observer detection. Although relative humidity can be associated with precipitation events, it does not explicitly capture rainfall occurrence. Because our goal was to account for broad-scale atmospheric conditions rather than model detailed weather dynamics, we focused on temperature and humidity as general indicators of environmental conditions that may influence bird detectability.

Statistical analysis

Our final dataset included detection/non-detection data for 84 study species from 98,960 unique eBird checklists submitted by 9,838 unique observers. To model the probability of observing each of the 84 study species during the breeding seasons of 2021–2023, we fit separate generalized linear mixed models (GLMMs) for each species using a binomial response variable. We modelled the probability of observing species i in checklist j as.

logit(p_{i, j}) = β₀ + β₁ log(PM_2.5j +1) + β₂Temperature_j + β₃RelativeHumidity_j + β₄LandCover_j + β₅Duration_j + β₆Distance_j + β₇Time_j + β₈Day_j + β₉Year_j + u _{observer (j)}.

where p_{i, j} represents the probability of detecting species i on checklist j. PM_2.5j represents the daily mean concentration of PM_2.5r associated with checklist j. We included the daily mean concentration of PM_2.5 as a fixed effect to evaluate how day-to-day changes in smoke pollution impacted detectability of birds. We also included fixed effects of daily mean air temperature and daily mean relative humidity to account for any effects of weather on detectability. To account for differences in habitat that may impact the presence and activity of birds, and therefore their detectability, we incorporated land cover as an additional fixed effect. We expected that shorter observation periods would result in detection of fewer species; thus, we included survey duration and distance traveled as fixed effects, following best practices for modeling eBird data. Bird activity varies throughout the day and across the breeding season (Hall and Ross 2007); therefore, we included time of day (Time_j; i.e., time observations started) and day of year (Day_j) as covariates to account for broad temporal patterns in detectability. We also included year (Year_j) as a fixed effect to account for interannual variation in bird populations, observer effort, and other environmental factors that may influence detectability independently of PM_2.5. We allowed the intercept β₀ to vary by unique observer as a random effect (u _{observer (j)}) to account for the differences in skill and experience among birders. All environmental predictors were modeled as fixed effects.

Prior to model construction, we evaluated multicollinearity among all continuous predictor variables (see Supplementary Table 1). No variable pairs showed high correlation (all |r| < 0.7), indicating that multicollinearity was not a concern. We also examined the distributions of all continuous environmental predictors. As expected, PM_2.5 was heavily right-skewed, so we applied a log transformation [log(PM_2.5 + 1)] to normalize its distribution. All continuous variables were then standardized (mean = 0, SD = 1) to facilitate coefficient comparability. All resulting coefficients are presented on the logit scale; positive values indicate increased probability of detection, while negative values indicate decreased probability.

To evaluate the contribution of PM_2.5 as a predictor of bird detectability, we compared the full model (including PM_2.5) to a reduced model that excluded the PM_2.5 term but retained all other covariates. We then used Akaike Information Criterion (AIC) to compare model fit between the models with and without PM_2.5, with lower AIC values indicating better relative model fit. We also calculated conditional and marginal R-squared values to determine the extent to which our models accounted for variation in the species-specific probability of observing birds. To summarize how continuous predictors influenced detectability across species, we extracted the fixed-effect coefficient estimates for each continuous predictor from all the species-specific GLMMs. For each predictor, these species-specific coefficients were then summarized across models using violin plots with embedded boxplots (See Results; Fig. 7). Our goal was not to estimate a binary treatment effect of a single wildfire event, but rather to quantify how bird detection probability varies continuously across the full range of observed PM_2.5 concentrations, with repeated smoke incursions in June–August 2023 contributing to the upper tail of observed exposure values.

To examine whether species responses to PM_2.5 varied according to ecological strategy, we classified each species by foraging guild and primary habitat association using trait information from the AVONET database (Tobias et al. 2022). Species were grouped into foraging guilds based on their dominant feeding strategy and into habitat categories based on their primary breeding habitat. Specifically, we used the Trophic Level and Primary Lifestyle fields to group species into broad foraging guild categories. Because many bird species exhibit mixed diets, these guilds represent dominant dietary strategies rather than exclusive feeding behaviors. Several AVONET dietary categories contained relatively few species in our dataset; as such, we grouped these species into broader foraging guilds (omnivore, herbivore/granivore, insectivore, and aerial insectivore) to facilitate ecological interpretation and ensure sufficient sample sizes within trait categories for comparison. These classifications were used to summarize and visualize patterns in PM_2.5 responses across ecological groups.

Our information-theoretic approach revealed that the full model, which included PM_2.5 as a predictor, had lower AIC values than the reduced model for 59 of the 84 study species (70%) (Supplementary Table 3). Of these 59 species, 55 (93%) exhibited a statistically significant relationship between PM_2.5 and detection probability (Supplementary Fig. 1 and Table 3). Among these 55 species, 40 (73%) showed a negative relationship with PM_2.5, meaning the probability of observing these species decreased as the concentration of PM_2.5 increased (Fig. 4). The remaining 15 species (27%) demonstrated a positive relationship with PM_2.5, meaning the probability of observing them increased with PM_2.5 exposure (Fig. 4). Species for which PM_2.5 did not significantly influence detection probability based on model comparison and coefficient significance are presented in Table 1. Complete model results are shown in Supplementary Table 4.

To evaluate whether species responses to PM_2.5 varied by ecological strategy, we extracted the estimated PM_2.5 coefficients from the species-specific GLMMs and summarized their distributions across foraging guild and habitat categories (see Methods) using trait data derived from the AVONET database (Tobias et al. 2022) (Figs. 5 and 6). This grouping reveals that aerial insectivores and some wetland-associated species tend to exhibit neutral or positive PM_2.5 responses, whereas forest-dwelling insectivores tend to show negative associations (Figs. 5 and 6). Among foraging guilds, non-aerial insectivores showed the most consistently negative coefficient estimates, whereas aerial insectivores were distributed around or above zero (Fig. 6). Similarly, forest- and shrubland-associated species showed predominantly negative coefficients, whereas wetland and open-habitat species showed more variable responses including several positive associations (Fig. 5). The list of species included in each foraging guild and habitat category is provided in Supplementary Table 2.

As expected, we found that weather and survey conditions influenced the detectability of a range of species. The effects of air temperature and relative humidity differed among study species, reflecting variation in species-specific behavior and activity patterns (Fig. 7). Such variation likely arises because weather conditions can influence bird vocal activity, movement patterns, and observer detection ability. Survey duration consistently emerged as a strong positive predictor of detectability for nearly all species, indicating that longer observation times increase the probability of detection. The only exception was the Laughing Gull (Leucophaeus atricilla), whose detectability remained unaffected by survey length. Similarly, distance traveled during surveys positively influenced detection probability for most species, likely reflecting greater spatial coverage and higher encounter rates. Time of day was predominantly a negative predictor of detectability, with the probability of observing birds declining as the day progressed for 75 out of 84 species. This pattern aligns with known diel activity patterns, where many birds are more active and vocal in the early morning (Singer et al. 2025). Additionally, day of year, our proxy for phenological timing, significantly influenced detectability for 71 species, indicating that seasonal migration and breeding behavior shape species presence and activity in ways that influence the visual and auditory cues relied on by observers. Although, we modeled day of year and time of day as linear fixed effects, which captures the dominant directional trends in detectability across the breeding season and diel cycle. We acknowledge that these relationships may be non-linear in nature; future analyses should consider incorporating quadratic terms for these variables to more fully account for their influence on detection probability.

To assess whether spatial variation in background PM_2.5 could confound our findings, we evaluated whether species responses to PM_2.5 were associated with land-cover variables. Because urban areas consistently exhibit higher background PM_2.5 levels, species that frequently occur in human-modified landscapes could exhibit positive associations with PM_2.5 simply due to co-occurrence with persistently polluted locations rather than wildfire smoke exposure. To examine this possibility, we calculated correlations between species-specific PM_2.5 coefficients and species’ associations with the NLCD land-cover categories assigned to checklist locations, as described in the Methods section. PM_2.5 effects showed moderate correlations with human-modified (r = 0.37) and cultivated land cover (r = 0.31), but only weak correlations with forest cover (r = 0.17). These results suggest that while spatial variation in background pollution may contribute to some observed patterns, it does not fully explain the species responses to PM_2.5 observed in our analysis. These findings support interpretation of our findings as a smoke-driven signal, though future studies could benefit from standardizing PM_2.5 within ecological zones or incorporating fire-attributed pollution datasets (e.g., satellite-derived fire-specific PM_2.5) to better isolate temporal smoke impacts from spatial pollution baselines.

To evaluate whether species with low detection prevalence showed attenuated PM_2.5 effects, we examined the relationship between prevalence and absolute effect size across all 84 species (Supplementary Fig. 3). We found no evidence of power-driven attenuation (Spearman ρ = −0.20, p = 0.066), suggesting observed patterns are unlikely to be driven solely by reduced statistical power.

Our study reveals that wildfire smoke pollution altered participatory science observations of birds in New York State during the breeding seasons of 2021–2023. We further show that PM_2.5, a marker of wildfire smoke, measurably influences bird detectability in ways that vary systematically with ecological traits. Negative responses were particularly common among migratory forest songbirds, including warblers (Setophaga spp., Geothlypis trichas), thrushes (Catharus spp.), and vireos (Vireo spp.), whereas aerial insectivores and some wetland-associated species showed neutral or positive associations with PM_2.5. These patterns suggest that the ecological context in which a species is detected plays a central role in determining its vulnerability to smoke-related detectability changes, with important implications for how biodiversity monitoring programs interpret population trends during periods of elevated air pollution.

We observed clear ecological patterns in the magnitude and direction of PM_2.5 effects on bird detectability when grouping species by habitat and foraging guilds. Forest- and shrubland-associated species showed the most consistently negative PM_2.5 coefficients, whereas wetland and open-habitat species showed more variable responses, including several positive associations (Fig. 5). At the species level, detection of cryptic species such as the Blackpoll Warbler (Setophaga striata), Black-throated Green Warbler (Setophaga virens), Black-and-white Warbler (Mniotilta varia), Black-throated Blue Warbler (Setophaga caerulescens), Magnolia Warbler (Setophaga magnolia), and Chestnut-sided Warbler (Setophaga pensylvanica) declined significantly with increasing levels of PM_2.5 (Fig. 4). These species are typically small-bodied, visually and acoustically cryptic, and detected primarily through vocalizations, traits that may render them particularly susceptible to acoustic and visual disruption associated with wildfire smoke (Sanderfoot et al. 2024; Simamora et al. 2026). Our results align with prior evidence that wildlife vocalization decreases during and after smoke disturbance (Cheyne 2008; Erb et al. 2023; Lee et al. 2017; Sanderfoot et al. 2024; Simamora et al. 2026), and that smoke-induced atmospheric changes can mask bird calls and lower acoustic detection rates (Duarte et al. 2021; Gasc et al. 2018).

Foraging guild analysis revealed a complementary pattern. Non-aerial insectivores were most consistently negatively affected by PM_2.5, whereas omnivores and aerial insectivores demonstrated more variable or even positive responses (Fig. 6). Aerial insectivores forage in open airspace and are frequently detected visually, meaning the acoustic pathways through which smoke may suppress detection are less relevant for this group. Additionally, previous studies suggest that disturbance events including wildfire and associated smoke can alter insect activity and availability, potentially increasing aerial prey under certain conditions and temporarily enhancing foraging activity by aerial insectivores (Doherty et al. 2022; Hradsky et al. 2017; Liu et al. 2022). Shifts in predator-prey dynamics and other mechanisms by which smoke could influence detection of birds likely vary across systems and remain an important area for further study.

These findings have direct relevance for long-term avian monitoring programs. Many North American monitoring efforts, including the Breeding Bird Survey and Monitoring Avian Productivity and Survivorship (MAPS), are concentrated in spring and summer months when wildfire smoke events are increasingly frequent (EESC 2019; WERC 2017). Our results indicate that smoke should be treated as an additional source of variation in detection probability when analyzing monitoring data, analogous to established covariates such as weather and survey conditions. Failure to account for PM_2.5 could lead to spurious apparent declines in forest songbird detections during smoke years, or to apparent increases in aerial insectivore detections that may reflect behavioral shifts rather than genuine population changes.

Our findings both align with and extend prior work from the western U.S., where Nihei et al. (2024) documented declines in mist-net capture rates during acute smoke events, and Sanderfoot and Gardner (2021) reported variable responses in bird detectability under elevated PM_2.5 conditions, including positive associations for some forest-dwelling species. While our results agree in showing aerial insectivores as less negatively affected (Fig. 6), we find stronger and more consistent negative responses among forest-dwelling warblers than documented in western systems. For example, Black-throated Blue and Magnolia Warbler exhibited strong negative associations with PM_2.5 (Fig. 4), contrasting with the more neutral or positive responses observed in some western systems. This divergence likely reflects differences in ecological context, including vegetation structure, species composition, or historical exposure to wildfire smoke. Western landscapes are more frequently exposed to wildfire and may support species that are more adapted to smoke-related disturbance, whereas the landscapes in the northeastern United States are less adapted to recurrent smoke exposure. Additionally, smoke incursions in New York State in June–August 2023 was temporally acute but spatially extensive, which may have led to short-term behavioral responses, such as reduced vocal activity or temporary displacement, thereby lowering detectability. Differences in sampling structure and observer dynamics between studies may also contribute to these contrasting patterns.

We caution that some patterns, particularly positive PM_2.5 associations for wetland and aerial insectivore species, may partly reflect changes in birder behavior rather than bird behavior. During periods of poor air quality, observers may preferentially conduct surveys in open wetland habitats or from roadsides rather than undertaking longer forest walks, independently inflating detection probabilities for open-habitat species. Although our effort covariates and observer random effects partially account for this, these variables cannot fully account for systematic shifts in where or how people bird during smoke events. Future studies could examine changes in checklist protocol type and habitat submitted during smoke events as proxies for observer behavioral responses, or incorporate structured survey designs that hold observer location constant during changing air quality conditions (e.g., Project Phoenix). Our habitat classification assigned a single dominant land cover class to each checklist location; future work should explore proportional land cover at multiple spatial scales as a more nuanced characterization of survey habitat (Ramesh et al. 2022; Callaghan et al. 2018).

Our study demonstrates that wildfire smoke – even at distances far removed from active fire measurably alters the detectability of breeding birds – in ways that are predictable from ecological traits. As wildfire smoke events become more frequent and geographically expansive, integrating air quality data into biodiversity monitoring frameworks will be increasingly critical for accurately interpreting population trends and avoiding smoke-driven bias in long-term datasets.

Wildfire smoke reduces the probability of detecting birds during the breeding season in New York State, indicating that elevated particulate pollution can influence how birds are observed during monitoring surveys. These results suggest that bird monitoring programs should account for smoke conditions when interpreting survey data. In regions increasingly affected by wildfire smoke, incorporating real-time air quality information (e.g., PM_2.5 concentrations) into survey planning or detection models may help reduce bias in estimates of species occurrence and abundance during smoke events.

Patterns across ecological guilds further suggest that species’ foraging strategies and habitat associations may influence how detectability responds to smoke exposure. Understanding these differences will be important for interpreting survey results collected during periods of degraded air quality. More broadly, the increasing frequency of large wildfire smoke events across North America highlights the need to better understand how air pollution influences wildlife monitoring data.

Citizen-science datasets such as eBird provide an important opportunity to examine how environmental disturbances affect bird detectability across large spatial and temporal scales. Future research integrating behavioral observations, acoustic monitoring, and demographic data will help clarify how smoke exposure influences bird activity and habitat use, providing a more complete understanding of how wildfire smoke affects avian communities.

Authors

1. Festus O. Adegbola
2. Stuart M. Evans
3. Olivia V. Sanderfoot
4. Adam M. Wilson

Cite this article

O. Adegbola, F., M. Evans, S., V. Sanderfoot, O. et al. Wildfire smoke alters observations of 65% of breeding bird species in New York State. Biodivers Conserv 35, 204 (2026). https://doi.org/10.1007/s10531-026-03406-9

Download citation

• Received: 11 August 2025

• Revised: 19 June 2026

• Accepted: 20 June 2026

• Published: 03 July 2026

• Version of record: 03 July 2026

• DOI: https://doi.org/10.1007/s10531-026-03406-9

Keywords