Weather radars are increasingly used to study the spatial-temporal dynamics of airborne birds and insects. These two taxa often co-occur and separating their contributions is crucial for reliable interpretation of their movement patterns. Most studies have restricted analyses to locations, seasons, and periods in which one or the other taxa dominates. In this study, we describe an analytical method to estimate the proportion of birds and insects in cases where both taxa share the same airspace. Our approach partitions vertical profiles of biological reflectivity into bird and insect components, using assumptions of downwind heading selection by insects and information on expected airspeeds for birds and insects. We evaluated our method in regions, where existing approaches of studying bird migration with weather radars can be particularly challenging due to high airborne insect density: the tropics (Colombia) and the southern temperate zone (Southeast Australia). We found that bird and insect signals routinely reached similar magnitudes in these regions. Retrieved patterns of bird and insect occurrence across daily and annual cycles reflected expected biological patterns that are indicative of migratory and non-migratory movements in both climates and migration systems, particularly broad-front migration in birds. Contrary to fixed airspeed thresholding, we were able to partition birds along the full range of bird-insect proportions, retaining more spatial-temporal complexity that was crucial to revealing aerial habitat use of both taxa. Our analytical procedure readily extends existing vertical profiling approaches, empowering ecologists to explore complex aerial ecosystems across a diverse range of climates, as well as potential diurnal movements of birds and insects that remain heavily understudied. 

We obtained Colombian weather radar data including both single polarization moments (radial velocity and reflectivity) and dual polarization moments (correlation coefficient, differential reflectivity and differential phase) from the country's Institute for Hydrological, Meteorological, and Environmental Studies (IDEAM), focusing on the stations in San Jose del Guaviare, Guaviare department, south-central Colombia  (name: GUA, 2.53°N, longitude: 72.62°W, elevation: 218 m a.s.l.) and Barrancabermeja, Santander department, north-central Colombia (name: BAR, latitude: 6.93°N, longitude: 73.76°W, elevation: 80 m a.s.l., Fig. 2A). Both radars are dual-polarization and operate at C-band (5.3 cm) with a beamwidth of 1°. The radar operated in dual-PRF mode with a low pulse repetition frequency (PRF) of 1266.667 Hz and high PRF of 1900 Hz, corresponding to a Nyquist velocity of 50.6 m/s. For the Australian case study, we obtained the level-1 dataset from the Australian National Computational Infrastructure (NCI, Soderholm et al. 2019). We used data (radial velocity and reflectivity) from one radar in Captains Flat, New South Wales, southeast Australia (name: CapFlat, 35.66°S, longitude: 149.51°E, elevation: 1382 m a.s.l., Fig. 2B) for the year 2018. This is an S-band (10.4 cm wavelength) single-polarization radar with a beamwidth of 1.9°. Reflectivity data in the Australian dataset was calibrated by matching collocated information with the Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Measurement (GPM) satellite passes (Louf and Protat 2023) before further processing. The radar operated in dual-PRF mode with a low PRF of 500 Hz and a high PRF of 750 Hz, corresponding to a Nyquist velocity of 39.1 m/s.

The Colombian radars are located in hot tropical lowlands, where seasonality is defined more by rainfall than temperature. The Barrancabermeja radar lies within the Magdalena River Valley but its beams reach the slopes of the Andes. San Jose del Guaviare sits at the junction between the Amazon and Orinoco River basins. We chose these Colombian radars based on their low elevations that maximize the volume of airspace captured by a single scan, the continuity of the available time series, and their strategic importance for regional migration, ensuring a sufficient volume of migration capturable by the radars (Bayly et al. 2018). The CapFlat radar is the highest radar in Australia in terms of elevation, with winter temperature (average low in July) reaching around 1.1 ℃ and summer temperature (average high in January) reaching around 27.9 ℃. Combined with its southerly latitude, it is most likely to record seasonal patterns in biological activity aloft in the temperate regions of Eastern Australia.

We extracted vertical profiles of biological reflectivity (η, cm²⋅km^-3) using the vol2bird algorithm from the bioRad R package (Dokter et al. 2019) to a temporal resolution of 6 min and 10 min and an altitudinal resolution of 200 m and 100 m in the Australian and Colombian datasets, respectively. Precipitation in the Australian dataset was removed using MistNet, a convolutional neural network and screened manually by visually checking time-altitude plots of reflectivity factor and removing time-altitude ranges with meteorological contamination (Lin et al. 2019, Dokter et al. 2019). Precipitation in the Colombian dataset was removed using the correlation coefficient ρ_HV (pixels with ρ_HV > 0.95 were removed, Stepanian et al. 2016). To reduce the risk of precipitation contamination, an additional 5 km margin was removed around identified precipitation areas for all radars. The vertical profile includes the reflectivity and the associated U (east-west) and V (south-north) components of ground speed. We retrieved the U and V components of wind speed from the ERA5 reanalysis for Australia (Hersbach et al. 2023) and the North American Regional Reanalysis for Colombia (Mesinger et al. 2006). We downloaded the hourly (Australia) and 3-hourly (Colombia) data at pressure levels from 1000 to 750 hPa, spatially interpolated to the radar location, and further interpolated to the altitudinal and temporal resolution of the vertical profiles. Airspeed was determined by subtracting the wind speed vector from the ground speed vector. We found airspeeds in our case studies were skewed heavily towards lower values (Supplementary Material Fig. S1 and S2), even during heavy bird migration, because of the strong background of aerial insects.

To summarize daily and seasonal patterns, we calculated the mean bird proportion , bird reflectivity  and insect reflectivity  at each time-altitude bin of the vertical profile spanning a data set of four years from the BAR radar in Colombia (n = 171,986 scans), and one year at the CapFlat radar in Australia (n = 105,120 scans). We chose the BAR radar because time series from this radar was more complete, and the quality of velocity data was higher. To illustrate the seasonal patterns, we created yearly trends of bird proportions, bird and insect reflectivity by fitting Generalized Additive Models (GAM, Wood 2017) using a quasi-Poisson distributional family and cyclic cubic regression splines, using bird proportions and the densities as response and day of year as predictor.