Tracking and modeling the movement of Queensland fruit flies, Bactrocera tryoni, using harmonic radar in papaya fields

Abstract

Determining movement parameters for pest insects such as tephritid fruit flies is critical to developing models which can be used to increase the effectiveness of surveillance and control strategies. In this study, harmonic radar was used to track wild-caught male Queensland fruit flies (Qflies), Bactrocera tryoni, in papaya fields. Experiment 1 continuously tracked single flies which were prodded to induce movement. Qfly movements from this experiment showed greater mean squared displacement than predicted by both a simple random walk (RW) or a correlated random walk (CRW) model, suggesting that movement parameters derived from the entire data set do not adequately describe the movement of individual Qfly at all spatial scales or for all behavioral states. This conclusion is supported by both fractal and hidden Markov model (HMM) analysis. Lower fractal dimensions (straighter movement paths) were observed at larger spatial scales (> 2.5 m) suggesting that Qflies have qualitatively distinct movement at different scales. Further, a two-state HMM fit the observed movement data better than the CRW or RW models. Experiment 2 identified individual landing locations, twice a day, for groups of released Qflies, demonstrating that flies could be tracked over longer periods of time.

Keywords: Directional movement; Field tracking; Fractal analysis; Hidden Markov models; Modeling; Movement simulation.

Figure 1

HR tagged Bactrocera tryoni flight tracks for Experiment 1 (induced movement). Colored arrows represent a series of 10–12 flights for a single tagged fly. When all flights were taken together (inset top right), flight directions were homogeneous showing no directionality (P = 0.078, Rayleigh test; P = 0.097, Hermans-Rasson test).

Figure 2

Combined turning angles of HR tagged Bactrocera tryoni for Experiment 1. Combined turning angles for all step-distances (concentration of 0.61) (A) and steps greater than or equal to 0.8 m (B) were non-random by both Rayleigh and Hermans-Rasson tests, showed no right-left bias, but indicate a pronounced bias towards moving within 90° left or right of the directly previous flight. Conversely, combined turning angles for steps under 0.8 m (C) were random by both Rayleigh and Hermans-Rasson tests, showing no directional movement bias.

Figure 3

Hidden Markov model of Bactrocera tryoni movements in Experiment 1. The distributions of step lengths (A) and turning angles (B) are shown for the two-state model. State 1 (shown in red) is comprised of fly movements with shorter step lengths (A) and more random turning angles (B) while state 2 movements are generally longer (A) and more show a greater propensity to maintain a directional heading over multiple steps (B). Grey bars show the proportion of fly movements for a given step length (A) or turning angle (B). Example flight tracks are shown in C and D with state switching illustrated.

Figure 4

Combined fractal dimensions from eighteen of the observed Bactrocera tryoni movement paths (two shorter paths were excluded). Fitting a discontinuous two-phase model showed a change point in the fractal dimension curve at a spatial scale of 2.48 m (black dotted lines show the two linear estimations). This suggests that Qflies in papaya fields move qualitatively differently at spatial scales < 2.48 (blue dots) and > 2.48 m (red dots). Linear regressions were performed on log transformed data.

Figure 5

Bactrocera tyroni flight step-distances for Experiment 1. Step-distances were categorized into 1 m intervals for this analysis.

Figure 6

Mean squared displacement (m) distances by number of consecutive movement steps for Bactrocera tyroni in Experiment 1.

Figure 7

Comparisons between Bactrocera tyroni field tracking data (A) and model simulations. Example simulations of 100 flies each taking 100 steps based on three movement models (B). Dotted black lines show circles containing 95% of all steps. Example simulations of 20 flies each taking 10 steps, roughly matching the field tracking data are shown for (simple) random walk (Brownian) (C), correlated random walk (D), and hidden Markov (E) models.

Figure 8

HR tagged Bactrocera tryoni flight tracks for Experiment 2 (natural movement). Colored arrows represent a series of movements for a single tagged fly. Black Xs show locations where the detected fly could not be visually identified. When all flights were taken together (inset bottom right), flight directions were not homogeneous showing directionality (P = 0.004, Rayleigh test; P = 0.018, Hermans-Rasson test).