EXPLORING RADAR AS A TOOL FOR STUDYING MIGRATORY BIRDS AND THEIR RELATIONSHIPS WITH DYNAMIC LANDSCAPES

Abstract

As a crisis-based discipline, conservation biology necessitates that we make timely management decisions to protect species based on the best available information. In effect, as new scientific tools become available, we must contend with simultaneously applying them in ways that provide novel insights and evaluating their limitations to assess the validity of those insights. The integration of radar with machine learning is one such tool that has revolutionized aeroecology, or the study of airborne organisms. Burgeoning methodologies leverage this integration, offering unique opportunities to understand how migratory birds are responding to large-scale environmental changes, such as urbanization and shifting light regimes. In this dissertation, my goal was to elucidate the promises and shortcomings of radar and machine learning as tools for informing migratory bird conservation management amid rapid global change. As a first step, I focused on validating observations from a local weather radar station. Weather radar systems have become a central tool in the study of nocturnal bird migration. Yet, while studies have sought to validate weather radar data through comparison to other sampling techniques, few have explicitly examined the impact of range and topographical blockage on sampling detection—critical dimensions that can bias broader inferences. In my first chapter, I assess these biases with relation to the Cheyenne, WY Next Generation Weather Radar (NEXRAD) site, one of the large-scale radars in a network of 160 weather surveillance stations based in the United States. I compared local density measures collected using a mobile, vertically looking radar with reflectivity from the NEXRAD station in the corresponding area. Both mean nightly migration activity and within night migration activity between NEXRAD and the mobile radar were strongly correlated (r = 0.85 and 0.70, respectively), but this relationship degraded with both increasing distance and beam blockage. Range-corrected NEXRAD reflectivity was a stronger predictor of observed mobile radar densities than uncorrected reflectivity at the mean nightly scale, suggesting that current range correction methods are somewhat effective at correcting for this bias. At the within night temporal scale, corrected and uncorrected reflectivity models performed similarly up to 65 km, but beyond this distance, uncorrected reflectivity became a stronger predictor than range-corrected reflectivity, suggesting range limitations to these corrections. Together, these findings further validate weather radar as an ornithological tool, but also highlight and quantify potential sampling biases. In my second chapter, I focused on using NEXRAD to study habitat transitions by migratory birds. During migration, birds regularly transition between terrestrial and aerial habitats. Yet, much of our understanding of migratory behavior is centered around either terrestrial habitat quality or atmospheric conditions separately, at relatively coarse temporal scales. I employed NEXRAD to study the dynamic drivers and relative importance of terrestrial, aerial, and sampling predictors as birds transition between the terrestrial and airspace boundary. I found that atmospheric conditions were consistently strong predictors of migration activity throughout the night, and across spring and fall seasons. Key sampling predictors, such as the time after local sunset, fluctuated throughout the night, with high importance shortly after sunset and diminishing importance in the middle of the night. Yet, terrestrial variables were not a strong predictor of nightly variation in migration activity. My results demonstrate that the importance of predictors of activity varies temporally, both within a single night and across seasons. These findings illuminate bird migration as a dynamic process, highlighting limitations and opportunities for employing weather surveillance radar to study transitions between terrestrial and aerial habitats. In my third chapter, I used NEXRAD to study migratory stopover in urban areas. Despite global expansion, the role of cities in macroecological processes remains understudied. Using radar estimates of migratory bird stopover across the U.S., I assessed urban landscapes' contributions to stopover and links to social demographics for 2,130 parks across 88 cities. Stopover hotspots disproportionately occurred on urban landscapes relative to land area, with nearly 50% of spring migration hotspots falling within Metropolitan Statistical Areas. The relationship between urbanization and stopover varied regionally, correlating negatively in eastern flyways and positively in western flyways. Finally, stopover was positively correlated with income but varied considerably, with many cities showing no effect or an effect in the opposite direction. This study highlights the significance of cities in a hemispheric-scale ecological process and demonstrate radar as tool for studying urban social-ecological interactions. Finally, in my fourth chapter, I investigated the effects of different light spectrums on birds in flight for the purpose of informing novel conservation management approaches. Artificial light at night (ALAN) has been shown to influence the behavior of migratory birds, yet how different light spectra modulate these effects is somewhat unclear. I conducted a field experiment across 31 nights at a remote site in northern Colorado using LED floodlights with white, red, amber, and blue lighting treatments during fall migration in 2023 and 2024. Using a vertically looking radar system, I quantified avian in-flight responses in migration traffic rate, flight height, and flight direction. I found that short wavelength, white light significantly reduced flight height, and this response was stronger than red, amber, or blue lights. Beyond providing insight into avian biology, my results could have implications for the conservation management of ALAN. Further, the ability to detect behavior changes from a small point source in a low-density migration system supports the notion that ALAN may be more pervasive than is often recognized. At its core, my dissertation is indicative of a broader shift in ecology and conservation science. Advances in remote sensing offer an opportunity to vastly expand the way we characterize social-ecological systems thereby diversifying the options we have for managing them. However, this “big data” approach must be validated and informed by local inference. My work emphasizes this point. As the integration of large datasets and machine learning become increasingly prominent in conservation biology, I urge the conservation community to explore their potential with creativity while remaining vigilant of the potential biases they may introduce.