Accuracy of swallow roost locations assigned using weather surveillance radar

Abstract

Weather surveillance radars (WSR) have been used to detect roosting aggregations of swallows since the 1950s. We provide the first quantitative assessment of the accuracy of roost locations derived from WSR images. We found 265 swallow roosts in WSR images from the Eastern US (east of 100oW) between June and September over 7 years (2010–2016). We quantified error in WSR-based roost locations of 72 of these roosts by comparing them to ground-truth locations. Purple Martins (Progne subis) formed 67 (93%) of ground-truthed roosts. Tree Swallows (Tachycineta bicolor) and Bank Swallows (Riparia riparia) formed the other 5 roosts (7%); all of which were in the northeastern U.S. (north of 41oN and east of 80oW). Magnitude of roost-location error was 2.94 kilometers (SD = 1.63 km, N = 72, range = 0–7.3). This error increased slightly, but significantly, with distance from the WSR (R2 = 0.13, P = 0.002). Location error was not influenced by wind speed. Directions of location errors (from confirmed to estimated location) were clumped in a southeasterly direction (x̅ = 128.9°, SD = 1.32, N = 72, P < 0.05). Similarly, prevailing wind direction (x̅ = 180.0°, SD = 1.28, N = 71, P < 0.01) tended to come from the south. Wind direction and direction to the nearest WSR were not correlated with direction of the location error. Magnitude and direction of error associated with using WSR images to assign geographic locations to swallow roosts is small. This error is largely unaffected by variation in distance and direction to WSR stations and prevailing wind conditions. These results enable future studies that seek automated approaches to roost identification and permit association with other environmental data at the appropriate level of spatial precision.