Sex ratios influence spatial occupancy and kinematic stability of Anopheles coluzzii mosquito swarms

Malaria mosquitoes reproduce in mating swarms. Previous studies have reported a pronounced activity peak in male mosquito swarms immediately following simulated sunset, typically lasting around 20 min. This peak represents the main swarm formation, where several mosquitoes concentrate above visual markers and maintain prolonged flight activity. However, most studies rely on laboratory setups with balanced or single-sex swarms, which do not reflect the male-biased sex ratios observed in the field.

In this study, we studied swarming behavior of male and female Anopheles coluzzii mosquitoes in five sex ratios (male-only 1:0, male-biased 3:1, balanced 1:1, female-biased 1:3, female-only 0:1) using three-dimensional infrared videography to quantify spatial structure of swarms and flight speed of individual mosquitoes. For each ratio, we analyzed changes in spatial arrangement and flight speed through time and between conditions.

Swarm volume varied following a quadratic trend ( $R^2=0.889$ ). As the proportion of females in the swarm increased, the volume of the swarm increased, ranging from 305.1 cm³ in male-biased swarms to 612.6 cm³ in female-only swarms. Mean flight speed also increased with female proportion, from $0.66$  m/s (1:1 balanced ratio) to $0.87$  m/s (0:1 female-only ratio), showing a moderate relationship with volume ( $R^2=0.504$ ). Swarm density and speed were negatively correlated, indicating that mixed swarms are not only smaller in volume but also exhibit higher track densities ( $R_{\text{Spline}}^2=0.712)$ suggesting tighter, slower swarms in male-rich groups. Furthermore, we used a Random Forest as an exploratory classifier to (1) identify which kinematic features most differentiate operational sex ratio (OSR) groups and (2) test, as a proof of concept, whether sex ratio can be inferred from kinematic signatures.

These results demonstrate the influence of sex ratio on swarm kinematics and support the use of machine learning for behavioral classification in mosquito ecology.