Brown Pelican (Pelecanus occidentalis) groundings on a 4-lane divided highway in south Texas: A comparative study using citizen science

Abstract

Bird collisions with human infrastructure, including wind turbines, buildings, and utility lines, are well documented and often linked to adverse weather conditions or structural interference with wind flow (Drewitt and Langston 2008). Migratory birds face elevated collision risk when precipitation low temperatures, or reduced visibility from infrastructure impair their ability to detect obstacles and when strong or shifting winds reduce their ability to control flight, and force birds into atypical flight paths (e.g., lower flight height in poor flying conditions (Langston and Pullan 2003). Similar interactions between wind and built structures have been implicated in collision and grounding events on bridges, including cases in south Texas, where changes in wind direction, downwash, and turbulence were factors in causing brown pelicans to land unintentionally on the roadway (Owens et al. 1990). Grounding events are of particular concern because Bird collisions with human infrastructure may cause direct mortality from initial impact or indirectly expose grounded birds to secondary hazards such as vehicle collisions. Brown pelicans (Pelecanus occidentalis) are large-bodied seabirds characterized by long, relatively narrow wings (aspect ratio ~ 9.8) and moderate-to-high wing loading (~ 58 N.m⁻²). These morphological traits facilitate efficient coastal gliding, soaring, and high-speed diving but impose significant constraints on takeoff performance (Pennycuick, 2008;Stokes and Lucas, 2021). Like other large soaring seabirds, pelicans typically orient takeoff into headwinds and may require extended surface runs to achieve the air speed necessary for lift generation, although takeoff remains possible under favorable, windless conditions (Owens et al., 1990).Wind conditions are critical for both takeoff and sustained soaring. Empirical and theoretical studies demonstrate that reduced wind speeds increase the energetic cost of takeoff, while turbulent airflow, wind shear, and abrupt directional shifts can disrupt lift production and compromise flight control (Norberg, 1990;Pennycuick, 2008;Uesaka et al., 2023). Consequently, wind strength and direction strongly influence movement patterns and habitat use in species with high wing loading and limited lowspeed maneuverability (Clay et al., 2020;Thorne et al., 2023). The restricted maneuverability and demanding takeoff requirements of large soaring birds heighten the risk of strikes and accidental groundings when they encounter linear infrastructure. Along coastal bridges in south Texas, brown pelican groundings are most frequent during strong winter (Oct to March) cold fronts accompanied by rain or mist, when turbulence and "wind shadows" created by bridge structures force birds downward onto roadways (Owens et al., 1990;1991). Under these conditions, paved surfaces may function as ecological traps; they appear visually suitable for landing but lack the aerodynamic conditions or surface properties required for successful takeoff, increasing the likelihood of bird-vehicle collisions. This vulnerability is not unique to P. occidentalis as comparable risks have been documented in other pelican species, (e.g., Great White Pelican (P. onocrotalus) and the Dalmatian Pelican (P. crispus), and other large soaring taxa like White Storks (Ciconia ciconia), cranes (Grus spp.), and vultures (Gyps sp., Aegypius monachus) (Orłowski, 2005;Møller et al., 2011;Morales-Reyes et al., 2017). These taxa share traits (e.g., high wing loading and a reliance on wind-assisted flight (gliders)) that increase their susceptibility to grounding in turbulent airflow encountered near infrastructure (Orlowski 2005;Møller et al., 2011, Morales-Reyes et al., 2017). From 2016 to 2022, episodic brown pelican groundings and mortalities have occurred at the Carl "Joe" Gayman Bridge (hereafter Gayman Bridge) on State Highway 48 (SH 48), adjacent to the Bahia Grande (a 2,630-ha restored tidal basin) and the Brownsville ship channel (Fig 1 .; Birt et al. 2021). While pelican mortalities have been documented at the nearby Queen Isabella Causeway due to birds landing on the roadway and being struck by vehicles during routine movements (Shafer & Jasek, 1997), the mass-mortality events observed at the Gayman Bridge on SH 48 appear to result from a unique convergence of geography and pelican behavior. Pelicans navigating local waterways must cross the Gayman Bridge to reach isolated evening roosts on small islands. The east-west orientation of the span at the Gayman Bridge requires returning birds to fly north across the bridge. When strong cold fronts approach, pelicans are therefore forced to fly directly into powerful headwinds to reach these roosts, aligning peak evening movements with the most hazardous aerodynamic conditions (Birt et al., 2023).Similar but unpublished events have been observed at the Packery Channel Bridge in Corpus Christi, Texas (Birt et al., 2023), suggesting that pelican groundings and mortalities can occur at other coastal structures when bridge configuration, pelican movement patterns, and strong cold fronts intersect.During these episodic grounding events, brown pelicans attempting to cross the roadway lose lift and fall to the bridge surface, where they become vulnerable to vehicular collisions (Birt and Gelston 2018). Brown pelicans cross the bridge as part of their daily movements to and from roost sites on islands in the Bahía Grande and the Brownsville Ship Channel (Birt et al. 2021). Amplifying these risks, seasonal abundance in the region has been estimated to increase to 1,000-2,000 individuals during fall and winter and decrease to approximately 400 during summer (Birt et al. 2021).To diagnose why brown pelicans were becoming grounded, the Texas Department of Transportation (TxDOT) and the Texas A&M Transportation Institute (TTI) conducted wind-tunnel experiments replicating the roadway and bridge deck. These studies found that the 91.4-cm-tall Single-Slope Traffic Rail (SSTR) concrete traffic barriers (CTBs) along the roadway edges produced downwash over the westbound lanes, generating turbulent airflow that resulted in brown pelicans becoming grounded and unable to regain flight (Birt and Gelston 2018;Birt et al. 2023). TxDOT and TTI subsequently evaluated alternative railing types to determine which configuration would reduce wind turbulence while meeting traffic safety standards. Testing identified the 112-cm-tall T2P railing as the design that minimized turbulence and lowered the altitude of barrier-disturbed air while maintaining safety standards (Fig. 2A-B; Birt et al. 2021).TxDOT replaced the existing SSTR CTBs on both sides of the Gayman Bridge with Type T2P railing between August 2019 and January 2020. Replacement occurred along approximately 850 m on the north (lagoon-facing) side and 100 m on the south side, for a total of approximately 950 m of outer barrier modification. Median barriers were not altered. The Gayman Bridge span itself is approximately 76 m in length and lies within a 1.9-km segment of SH 48 bordering the Bahía Grande tidal basin. Thus, prior to replacement, approximately 950 m of outer bridge and adjacent causeway edge were under the SSTR configuration; following replacement, that same 950-m segment was under the T2P configuration. Barrier comparisons in this study therefore evaluate grounding responses relative to a defined and constant linear exposure length (~950 m) along the bridge and immediately adjacent elevated causeway where pelican groundings were concentrated.Here, we evaluate the influence of weather variables and railing type on grounding events along SH 48 to better understand how pelican groundings can be predicted and mitigated. Two key hypotheses were tested:(i) the daily odds of brown pelican grounding presence would decrease under T2P railing relative to SSTR and increase with higher daily wind speeds (average, minimum, maximum, and range), greater daily changes in wind direction, lower daily air pressure (average, minimum, maximum, and range), and lower daily air temperature (average, minimum, maximum, and range); and (ii) observed daily grounding counts would decrease under T2P railing relative to SSTR and increase with higher daily wind speeds, greater daily changes in wind direction, lower daily air pressure, and lower daily air temperature. Vehicle road mortality survey methods are described in detail in Beer et al. (2025). Briefly, weekly stop-and-exit (SE) road mortality surveys were conducted by vehicle at 64 km.h⁻¹ along SH 48 from 19 December 2016 through 28 February 2022. All surveys were conducted between 08:00 and 13:00. The roadway was surveyed in both directions, and the vehicle was stopped at each observed road mortality to identify the carcass and record location data. Carcasses documented in previous weeks were recorded as present; however, analyses in this study included only the initial record of each brown pelican mortality. The dataset included 163 brown pelican road mortalities during the study period.Counts of brown pelican groundings were collected by citizen scientists (CS) around the Gayman Bridge while they volunteered to rescue grounded pelicans. Records were compiled by the Coastal Bend and Bay Estuaries Program from 8 December 2016 (the date of their first recording) through 28 February 2022. The CS observations occurred both incidentally as volunteers passed by and on specific dates they mobilized to rescue grounded pelicans. Mobilization occurred whenever citizen scientists predicted that brown pelican groundings would be plausible given weather conditions. The number and identity of citizen scientists varied among observation dates. However, mean observer effort did not differ between days when grounded brown pelicans were recorded (5.26 ± 0.69 observers) and days without groundings (4.96 ± 0.62 observers; Mann-Whitney U = 1043.5, p = 0.775).Citizen science observations occurred both on foot and in vehicles while approaching and leaving the Gayman Bridge. Because brown pelicans are large-bodied and visually conspicuous, substantial differences in detection probability between survey modes were unlikely. Observation dates differed in hours of the day spent observing brown pelicans but ranged from 06:30 hours at the earliest to 20:00 hours at the latest. For this study, "daily" was defined as the time range 06:00 hours through 21:00 hours, providing a 30-minute buffer rounded up to the nearest hour. This ensured that weather data only from times potentially relevant to brown pelican groundings would be used.For each date in the CS dataset, citizen scientists separately counted brown pelican mortalities and brown pelicans that were grounded but survived that were observed in the vicinity of the Gayman Bridge. Dead and grounded counts were combined in the CS dataset to give the total number of brown pelicans grounded on each date. The CS dataset included notes documenting inconsistencies in the timing of recorded pelican mortalities (e.g., mortalities recorded on one date but noted as having occurred on a previous date). These discrepancies were corrected by reassigning records to their appropriate dates.In contrast to the weekly road mortality surveys, citizen scientists actively removed grounded and recently deceased brown pelicans from the roadway. Mortalities were frequently moved off the roadway by CS, TxDOT personnel, or Texas Game Wardens and placed in piles beyond the travel lanes.Consequently, CS observations primarily reflected recent grounding and mortality events rather than long-persisting carcasses. As a result, daily CS counts more accurately represented the impact of the Gayman Bridge on brown pelican groundings. Because carcasses were routinely removed and records were reassigned when necessary to reflect the correct date of mortality, the likelihood of repeated counts of the same individuals across multiple days was minimal.Weekly road mortality surveys were known to be influenced by carcass removal between sampling intervals. Visual comparison of the weekly survey dataset and the CS dataset indicated that no major pelican mortality events detected in the weekly surveys were absent from the CS records (Figure 3). Therefore, analyses proceeded using the finer-scale (daily) CS dataset. Weather data were procured from the National Ocean Service station PTIT2 (26°3'40" N, 97°12'56" W) (National Oceanic and Atmospheric Administration 2022). Station PTIT2, located in Port Isabel, Texas, took readings of air temperature (C), air pressure (hectopascals (hPa), wind speed (m.s -1 ), gust speed (m.s -1 ), and wind direction (degrees clockwise from true north). Readings were reported every 6 minutes, with all but gust speed as averages over that interval. Gust speed was reported as the peak 5-8 second gust speed during the interval (National Oceanic and Atmospheric Administration 2022). This raw data over the study period (8 December 2016 through 28 February 2022) was aggregated and sets of recordings (all readings from one point in time) containing errors (e.g., 99 °C for air temperature) for any of the mentioned variables were removed. Three grounded brown pelicans that were recorded on 31 October 2019 were therefore excluded from further analysis as there was no error-free data available for that date.The corrected data were used to calculate daily averages, minima, maxima, and ranges of each type of reading except wind direction. The daily sum of the changes in wind direction between each reading was also calculated. All these calculated variables were visually compared to determine which ones to use in modelling (Supplementary Fig. 1). Gust and wind speed measures appeared to correlate and past analysis of fine-scale wind data did not reveal any evidence of wind gusts that could affect brown pelican mortality (Birt et al. 2021), so gust speed was not considered in analyses. Minima of air temperature, air pressure, and wind speed appeared to correlate with the maxima of each respective variable closely; therefore, measures relating to cold fronts were chosen because they were considered most relevant to modelling brown pelican groundings. These measures included the daily maximum wind speed, the daily minimum air pressure, and the daily minimum air temperature. The daily sums of changes in wind direction between recording, the month of grounding recording, and both the daily average and the daily range in air temperature, air pressure, and wind speed were selected for analysis as well. No grounded brown pelicans were recorded in the CS dataset between 2 March and 10 October in any year. To account for the possibility of missed groundings and given that road-mortality surveys detected pelicans within this broader seasonal window, a one-week buffer was applied at both ends of this interval. Consequently, dates from 9 March through 2 October (inclusive) of each year were excluded from the dataset. After applying these exclusions, the All-Groundings dataset comprised a total of 867 days.To examine the effects of barrier type, a subset of the All-Groundings dataset was created to exclude the construction period of the T2P railing (August 2019 through January 2020). This dataset, called Barrier Groundings, included 741 days. Grounded brown pelicans were observed on 36 of these days. SSTR rails were present on 405 of the 741 dates, and T2P railing was present on 336.To model the effect of different weather variables on grounded brown pelican counts, a dataset, called All-Counts, consisting only of days when citizen scientists were present at the study site, was used.A total of 49 dates were included. A subset of the All-Counts dataset was also created in which barrier type was included as a factor. This dataset, called Barrier Counts, only included the days without the construction period of the T2P railing (Aug 2019 through Jan 2020) (n = 42 days). Of the 42 dates included in the Barrier Counts dataset, 19 occurred under SSTR railing and 23 occurred under T2P railing. To model the different variables and the grounded brown pelican presence/absence, a binary logistic regression was used with the All-Groundings dataset. The variables included in modeling consisted of the daily sum of changes in wind direction between readings, the month of grounding recording, the daily maximum, average, and range in wind speed, the daily minimum, average, and range in air temperature, and the daily minimum, average, and range in air pressure. Linearity in the logit of the variables was tested using the Box-Tidwell procedure (Box and Tidwell 1962). Modeling began with the inclusion of all variables that passed the Box-Tidwell test and insignificant variables were eliminated in a stepwise fashion (Yamashita et al. 2007). Colinear variables (minimum or maximum, averages, and daily ranges within each category) were substituted for one another to obtain the best fit for each variable in each category in the model (Yamashita et al. 2007). The significant model with the lowest Finite Sample Corrected Akaike's Information Criterion (AICc) value was chosen as the top-ranked model (Sugiura 1978, Hurvich andTsai 1989). The same process was used with the Barrier Groundings dataset, including barrier type (SSTR or T2P) as an additional factor.Modeling of the All-Counts dataset began with inclusion of all weather variables described above.Variables that were colinear (e.g., daily minimum, maximum, mean, and range within the same weather category) were evaluated separately, and the best-fitting variable from each category was retained based on model fit. Insignificant variables were removed in a stepwise fashion using AICc model selection.The count response variable was initially modeled using Poisson regression. However, because the variance exceeded the mean (indicating overdispersion), the assumptions of equidispersion were violated (Lovett and Flowerdew 1989). Accordingly, negative binomial regression was used (Hilbe 2012).The model with the lowest AICc value was selected as the top-ranked model. The same procedure was applied to the Barrier Counts dataset, with barrier type (SSTR or T2P) included as an additional categorical predictor. Grounded brown pelicans were recorded on 42 of the 867 days used. The top-ranked statistically significant binary logistic regression model using the All-Groundings dataset had an Akaike weight value (wi) (Wagenmakers and Farrell 2004) of 0.8609 (P < 0.001) (Table 1). The best supported model indicated that the odds of grounding increased on colder, lower pressure, and windier days (Table 2). The likelihood of pelican grounding decreased with increasing daily minimum air temperature (0.817 times as likely per 1 °C increase, 95% CI [0.760, 0.878], P < 0.001) and with increasing daily average air pressure (0.894 times as likely per 1 hPa increase, 95% CI [0.834, 0.958], P = 0.008), and increased with increasing daily maximum wind speed (1.657 times as likely per 1 m.s⁻¹ increase, 95% CI [1.426, 1.925], P < 0.001) (Table 2). Using the reduced Barrier Groundings dataset (construction period excluded) the model including daily minimum air temperature, daily average air pressure, and daily maximum wind speed as covariates and barrier type as a factor was significant (P ≤ 0.001). However, barrier type was not a significant predictor of occurrence (SSTR vs T2P, P = 0.735), indicating that meteorological conditions, rather than barrier characteristics, primarily drive the odds of encountering a grounded pelican on a given day. Citizen scientists recorded a total of 1,004 grounded brown pelicans during the 49 survey days with an average of 20.5 ± 41.1 grounded brown pelicans per survey day (mean ± 1 SD). The count data exhibited overdispersion (variance exceeding the mean), violating the assumptions of a Poisson model. Accordingly, negative binomial regression was used. Daily average and minimum air pressure were excluded from the final model due to collinearity with other weather predictors.The top-ranked negative binomial regression model (wi = 0.3656, P ≤ 0.001) for grounded brown pelican counts (Table 3) showed an association with daily minimum air temperature (P < 0.01), daily average air pressure (P < 0.01), and barrier type (P < 0.001) (Table 4). Brown pelican groundings decreased by 80.7% (95% CI [60.8%, 90.5%]) with T2P railing compared to SSTR and declined by 12.0% (95% CI [3.4%, 19.7%]) per 1 hPa increase in daily average air pressure and by 9.4% (95% CI [3.7%, 14.9%]) per 1°C increase in daily minimum air temperature. Linear trendlines of the predicted mean brown pelican groundings show consistent negative trends versus daily minimum air temperature, daily average air pressure, and T2P vs SSTR (Fig. 4). Our findings demonstrate that replacing the SSTR with T2P railing at Gayman Bridge reduced pelican groundings with a decrease of 60.8-90.5%, and an average reduction of 80.7% in grounded counts. This stands in contrast with a previous assessment by Birt et al. (2021), who concluded that the T2P railing had not significantly reduced pelican grounding events. However, they cautioned that less than two years of post-installation data were likely insufficient to discern long-term effects given the inherent variability in cold-front strength, pelican movements, and observer effort. Our additional two years of monitoring show clearer evidence that the T2P retrofit substantively decreased the severity of grounding events, even though it does not eliminate them.The strongest predictors of grounded pelican presence were higher daily maximum wind speeds, lower daily minimum air temperature, and lower daily average air pressure, consistent with conditions characteristic of strong winter cold fronts in South Texas. Our results are consistent with previous analyses of brown pelican grounding/mortality events in the Bahía Grande, showing that major grounding events coincide with cold fronts (Birt et al. 2021). Rapid wind direction changes and post-frontal north winds generate "strong headwinds" for brown pelicans crossing to evening roosts in the Bahía Grande, and the proportion of time the wind direction is between 275 and 335 degrees was previously linked to daily mortality risk (Birt et al. 2021). Incorporating directional statistics or circular measures of wind change could further improve model performance.The dominance of meteorological factors in our models suggests that cold-front dynamics trigger the aerodynamic conditions necessary for pelican groundings to occur, irrespective of barrier type.Although barrier type did not significantly affect whether at least one grounding occurred on a given day, it had a strong effect on the number of pelicans grounded. Grounding occurrence reflects broadscale exposure; when a strong cold front hits at least a few pelicans are forced low regardless of railing type.However, the severity is strongly influenced by how the barrier disrupts air flow across the travel lanes.Wind-tunnel experiments of a model Gayman Bridge (Birt & Gelston 2018;Birt et al. 2023) demonstrated that SSTR rail create a steep downwash and deep turbulent wake, whereas T2P railing generates a shallower, higher-altitude disturbance zone. Our model results indicate this is sufficient to prevent mass grounding events, even when meteorological conditions predispose birds to forced landings. Thus, the T2P retrofit mitigates the number of birds compromised during hazardous events without altering the fundamental meteorological triggers underlying grounding occurrence.Brown pelicans intentionally exploit ground effect when soaring close to waves, achieving a 15-25% energy savings at average altitudes of 33 ± 5 cm (1 SD) (Hainsworth 1988, Stokes andLucas 2021).Maintaining energy efficiency may be a factor in explaining why brown pelicans remain close to the water surface prior to rising to fly over the Gayman bridge even though it may increase their susceptibility to unexpected turbulence and risk of being grounded. Similarly, their "dropping precipitously to the lagoon surface" (Birt et al. 2017) once past the bridge may be to regain ground-effect energy efficiency rather than an effect caused by the railing. Future research examining brown pelicans' approach to and exit over the bridge (including under differing wind conditions) may reveal whether modifying low-altitude flow conditions on one or both sides of Gayman Bridge could encourage higher flight levels.The channel leading to the Gayman Bridge was recently widened and may affect pelican movements and air flow patterns in ways not captured by the current dataset. In addition, the annual variation in pelican abundance (1,000-2,000 pelicans in winter; about 400 in summer; Birt et al. 2021) may also affect pelican grounding risk. In addition, continued monitoring is suggested to evaluate whether the present reduction in grounded counts persists as infrastructure, channel morphology, and the local pelican population evolve.We did not implement formal carcass persistence or detection probability corrections (e.g., capture-mark-recapture approaches; Guinard et al., 2012), which are sometimes used in road mortality studies to quantify detection and carcass persistence biases. However, brown pelicans are large-bodied and visually conspicuous, and mean observer effort did not differ between days with and without recorded groundings. In addition, citizen scientists actively removed grounded and recently deceased individuals from the roadway, reducing the likelihood of repeated counts across dates. Although unquantified variation in carcass persistence and detection probability may introduce some uncertainty into daily count estimates, these factors are unlikely to have systematically biased the relationships observed between weather conditions, barrier type, and grounding severity. Our findings show that while winter cold front conditions determine when brown pelican groundings occur at the Gayman Bridge, barrier design strongly influences how many birds are affected.Replacing concrete traffic barriers such as SSTR with T2P railing reduced grounded pelican counts by an average of 80.7%, demonstrating that aerodynamic rail profiles can mitigate mass grounding events even though some grounding events may still occur.Because risk is highest during periods of high winds, low temperatures, and low air pressure, forecasted cold fronts provide a practical early warning tool for managers. These conditions can be used to pre-position rescue teams, adjust traffic awareness, and coordinate monitoring during peak-risk periods. Since completion of this study, TxDOT has installed dynamic messaging signs near the Gayman Bridge that may be activated between October and April when winter cold fronts are forecasted.Given recent channel modifications at the Gayman Bridge, new traffic control devices, placement of utility lines, and continued expected pelican population growth, continued monitoring is warranted.However, current evidence supports T2P railing as an effective mitigation strategy and highlights the value of integrating weather-based forecasting with infrastructure design to reduce pelican mortality in South Texas. Collectively, these steps offer a practical and scalable framework that may serve as a model for mitigating seabird-infrastructure interactions in other coastal systems.More broadly, avian mortality on linear transport infrastructures reflects the interaction between intrinsic biological traits such as flight behavior, maneuverability, and aerodynamic strategy, and extrinsic environmental and structural features including land cover, habitat configuration, roadside vegetation, and infrastructure design (Orłowski 2008;Guinard 2013). In this system, winter cold fronts create aerodynamic conditions that interact with the brown pelican's low-altitude flight behavior and use of ground effect, increasing susceptibility to turbulence at bridge crossings. Our results demonstrate that modifying infrastructure aerodynamics can substantially reduce the severity of grounding events even when intrinsic behavioral tendencies and large-scale meteorological drivers remain unchanged. Integrating speciesspecific flight ecology with infrastructure and landscape design may therefore provide a transferable framework for reducing avian mortality across transport networks.