Fig 1.
Example flight patterns and illustration of variability.
A. Example fruit fly trajectory in a wind tunnel containing a stationary ethanol plume. The plume outline is shown in red, with the shell indicating 2 standard deviations from the plume’s centerline. The trajectory color shows the instantaneous odor concentration experienced by the fly, and the green star marks where the fly took off. B. Example plume-crossings from several trajectories. Blue arrows on a subset of crossings indicate the direction of flight. C. Heading is defined as the angle between the upwind direction and the fly’s velocity vector. D. Heading time courses surrounding example crossings (gray), as well as mean (black solid) and standard deviation (black dashed), relative to the crossing time. E. Odor concentration time courses surrounding example crossings.
Fig 2.
Dependence of heading on peak concentration.
A. Partial correlation of peak odor concentration experienced by the animal and subsequent heading at various times past the odor peak for fruit flies and mosquitoes. Shading indicates 2.5–97.5% confidence interval. B. Heading at 300 ms since odor concentration peak as a function of peak odor concentration during crossing. Each point represents a single plume-crossing. Though a significant correlation exists, the joint distribution is dominated by variability. C. Threshold and threshold-linear models for identifying non-binary dependence of h300 (the heading 300 ms post-crossing) on concentration. For mosquitoes, h500 (the heading 500 ms post-crossing) was used. P-values indicating the probability that the threshold-linear model would have fit better by chance (F-test) are shown in the gray box for four different experiments (fruit flies following an ethanol plume in 0.3, 0.4, or 0.6 m/s winds, and mosquitoes following a CO2 plume in 0.4 m/s wind).
Fig 3.
History dependence of crossing-triggered turns in data and models.
A-D. Crossing-triggered heading time courses for early (blue) or late (green) crossings for fruit flies tracking an ethanol plume in three different wind speeds, or mosquitoes flying in a wind tunnel with a 0.4 m/s wind. Thick lines indicate means, with shading indicating standard error of the mean. E. Same as A but for trajectories generated using the surge-cast model. F. Same as A but for trajectories generated using the centerline-inferring model. G-J. Same as A-D but for trajectories generated using the infotaxis algorithm, with wind speeds and plume profiles matched to each of the four experiments.
Fig 4.
History dependence of post-crossing turns after accounting for position and flight time.
Each panel shows the mean and standard error of the mean (shading) of h*(t) across different time points post-crossing for early (blue) and late (green) crossings. The dashed line indicates the probability that such a difference in means would have arisen from chance, with the gray line marking 0.05. A. Fruit fly experiment in 0.3 m/s wind (783 early crossings, 766 late crossings). B. Fruit fly experiment in 0.4 m/s wind (1158 early, 828 late). C. Fruit fly experiment in 0.6 m/s wind (394 early, 294 late). D. Mosquito experiment in 0.4 m/s wind (125 early, 143 late).
Fig 5.
Example empirical and simulated trajectories.
Each of the simulated trajectories (B-D) began at the same location in the wind tunnel (green star) as the empirical trajectory (A) and was run for the same length of time as the empirical trajectory.
Fig 6.
Analysis of hybrid surge-cast-infotaxis crossing model.
A. Hybrid surge-cast-infotaxis crossings, as described in text, separated into early vs. late crossings with a crossing model mixing 65% surge-cast with 35% infotaxis. B. The late vs. early difference in mean headings of the hybrid crossings (time-averaged from 0.3 to 0.8 s post-crossing) as a function of percentage of surge-cast making up the model (red), overlaid with that calculated from the 0.3 m/s wind speed fruit fly experiment (black). C. The distribution of data-matched surge-cast % over multiple random pairings of surge-cast and infotaxis crossings.
Fig 7.
Trajectory heatmaps for data and models.
Each distribution was calculated by binning all timepoints generated from all trajectories, empirical or modeled, used in our analyses. In this figure we have not excluded any timepoints based on position in the wind tunnel or heading. A. Empirical trajectories from fruit flies in a wind tunnel with an ethanol plume and 0.3 m/s wind. B-D. Equivalent trajectories generated by the surge-cast (B), centerline-inferring (C), or infotaxis (D) algorithms.