3D coordinates of the foraging flights of wild black skimmers

Birds commonly exploit environmental features such as columns of rising air and vertical windspeed gradients to lower the cost of flight. These environmental subsidies may be especially important for birds that forage via continuous flight, as seen in black skimmers. These birds forage through a unique behavior, called skimming, where they fly above the water surface with their mandible lowered into the water, catching fish on contact. Thus, their foraging flight incurs the costs of moving through both air and water. Prior studies of black skimmer flight behavior have focused on reductions in flight cost due to ground effect but ignored potential beneficial interactions with the surrounding air. We hypothesized a halfpipe skimming strategy for skimmers to reduce the foraging cost by taking advantage of the wind gradient, where the skimmers perform a wind gradient energy extraction maneuver at the end of a skimming bout through a foraging patch. Using video recordings, wind speed, and wind direction measurements we recorded 70 bird tracks over four days at two field sites on the North Carolina coast. We found that while ascending the skimmers flew more upwind and then flew more downwind when descending, a pattern consistent with harvesting energy from the wind gradient. The strength of the wind gradient and the flight behavior of the skimmers indicate that the halfpipe skimming strategy could reduce foraging costs by up to 2.5%.

Methods
The dataset was collected using videos from three digital Canon EOS 6D digital SLR cameras equipped with 50-millimeter lenses. Videos were recorded at 29.97 frames per second. They were synchronized by visual events identified separately in each camera. They were then calibrated for 3D position reconstruction following the protocol from Corcoran and Hedrick (2019) via bundle adjustment using previously determined pinhole model lens parameters and shared 2D information visible in at least two of the three cameras. The scene scale was established from the distances between the three cameras, and calibration parameters were converted to direct linear translation (DLT). The calibration was aligned to place the water surface at Z=0 with positive Z pointing upward and the X and Y directions forming the horizontal plane, with the X axis aligned to wind direction such that bird movements in the increasing horizontal X axis were downwind (in the direction of the wind) and decreasing was upwind (against the wind). Bird track coordinates were acquired with DLTdv (Hedrick, 2008), and bundle adjustment calibrations were performed using the MATLAB computer vision toolbox. Each bird track was then smoothed using a 4-pole digital Butterworth low pass filter with a cutoff frequency of 0.5 Hz.