Evaluating AI-based mosquito monitoring technologies: a field study in Zhejiang Province, 2025

This study aimed to assess the effectiveness of intelligent mosquito surveillance devices implemented in Zhejiang Province and to compare these metrics with those obtained from traditional surveillance methods.

Intelligent surveillance devices, light traps, and BG-traps were operated concurrently at the same locations. Statistical comparisons were conducted using one-way ANOVA with Dunnett's T3 tests, while the Dynamic Time Warping algorithm was employed to evaluate similarities in density trends.

A total of 278 trap-nights using light traps and 278 trap-hours of BG-traps captured 11,349 and 1,524 mosquitoes, respectively, yielding mean densities of 14.56 mosquitoes/(trap·night) and 6.34 mosquitoes/(trap-hour). Additionally, 991 and 646 trap-days of Chengwen Jingling traps and CAMMADE traps recorded 12,003 and 1,180 mosquitoes, respectively, based on backend data, with mean densities of 5.62 mosquitoes/(trap-day) and 1.83 mosquitoes/(trap-day). The light trap produced significantly higher mosquito densities than all other methods (P < 0.05), while the BG-trap exhibited comparable densities to Chengwen Jingling traps (P > 0.05) but significantly outperformed the CAMMADE trap (P < 0.05). During the same time window on the same day, the trapping efficacy of the BG-trap was significantly higher than that of the Chengwen Jingling Trap (P < 0.05), while no significant difference was observed between the light trap and the Chengwen Jingling Trap (P > 0.05). Temporal pattern analysis further revealed consistent trends among the Chengwen Jingling, light trap, and BG-trap methods. The Chengwen Jingling trap exhibited relative error rates (RE) of 25.67% for density estimation, 64.29% for Culex genus identification, and 179.20% for Aedes genus identification. The CAMMADE trap demonstrated RE of 105.34% for density estimation, along with 15.99% for Culex and 34.69% for Aedes at the genus level. For female mosquito identification, Chengwen Jingling achieved RE of 21.54% for Culex and 76.72% for Aedes, while CAMMADE showed rates of 102.99% for Culex and 153.85% for Aedes.

AI-based mosquito monitoring technologies exhibit significant potential for advancing vector surveillance; however, current technical limitations in trapping efficacy and algorithm accuracy necessitate a hybrid approach integrating AI with conventional methods in the near term, rather than a full replacement.