This dataset quantifies offshore and adjacent terrestrial migration of landbirds along the western North Atlantic and Gulf of Mexico coasts from 2014–2023. Offshore flights of migratory landbirds are common but difficult to quantify due to lack of offshore monitoring. Offshore wind development is expanding rapidly and may overlap with migration corridors. Understanding the timing, locations, and numbers of offshore birds can identify high-risk periods and areas for collision mitigation. We used 16 coastal weather surveillance radars in the western North Atlantic and Gulf of Mexico to evaluate differences in passage, altitude, and timing of migration between adjacent terrestrial and offshore habitats for spring and fall from 2014 to 2023. We used machine learning methods to predict unobserved low altitude densities that the radar beam overshoots at increasing distances from the radar. Migration traffic varied geographically and seasonally, with offshore migration traffic being lower in spring than fall and lower than terrestrial migration year-round. Offshore migratory activity below 300m (common rotor-swept zones) showed shorter durations and fewer peak nights than terrestrial migration. Offshore flight altitudes were 8-14% lower than over land, and the propensity for offshore flights was higher in fall, especially near coastlines perpendicular to migratory directions. We identify regional patterns in offshore migration to inform wind energy siting decisions and operation. Migration of landbirds in the offshore environment at rotor swept zones is common, and increasing wind energy operations will likely increase interaction between birds and rotors. The limited number of peak nights and shorter migration windows offshore increases opportunities for dynamic conservation. Curtailment targeting high-risk regions, seasons and nights, will reduce operational downtime while protecting birds. Fall migration and areas downstream of coastlines perpendicular to migratory direction have a higher propensity for offshore migration, suggesting that spatial and temporal variation in migration intensity should be considered when balancing collisions risks and operational costs. 

This data is vertically integrated vertical profiles of 16 coastal NEXRAD stations, which was filtered to isolate biological activity while filtering out precipitation and sea clutter noise. To compile this dataset, we analyzed bird migration data from 16 coastal WSR-88D (NEXRAD) radar stations along the Atlantic and Gulf coasts of the U.S. between 2014 and 2023 between the spring (Mar 1–Jun 15) and fall (Aug 1–Nov 15) migration seasons. We downloaded Level II polar volume data at 30-minute intervals between civil sunset and sunrise for each radar from the NEXRAD archives (Original polar volume files are from the NEXRAD Level II archive data which are openly available from the Registry of Open Data on AWS Data Exchange (https://registry.opendata.aws/noaanexrad).

Using the R packages bioRad and vol2birdR, we derived vertical profiles of bird density from reflectivity data, excluding precipitation via a correlation threshold (ρHV > 0.95). Each radar scan was partitioned into terrestrial and offshore profiles using custom spatial masks based on high-resolution coastline data, excluding a 10 km coastal buffer.

We calculated three types of vertical profiles using 100 m bins from 0 to 5000 m a.s.l.:

1. Standard (5–35 km range)
2. Terrestrial (5–75 km)
3. Offshore (5–75 km)

A post hoc precipitation filter and manual review removed remaining precipitation and sea clutter, excluding 435 radar-date combinations.

Because radar beams overshoot lower altitudes at greater ranges offshore, we developed an extrapolation model using xgboost to estimate densities in unsampled low-altitude bins (0-100 to 0-400 m) based on standard land-based profiles. Profiles were normalized by vertically integrated density and cube-root transformed to reduce skew. Four models were trained, each predicting an increasing number of lower bins, and performance was validated against radars with complete low-altitude coverage.

We calculated height-integrated bird density and flight altitudes using bioRad’s integrate_profile(), which converts densities into migration traffic rates (birds km^-1^h-1^). Temporal integration produced nightly and seasonal passage estimates.

This resulted in the attached file "Integrated_radar_profiles.csv" which was used in the analysis.