Mosquito surveillance still needs conventional traps, despite AI gains

AI-based mosquito monitoring is getting a real-world test, and the early verdict is mixed. In a 2025 field study published in Frontiers in Veterinary Science, investigators in Zhejiang Province compared intelligent mosquito surveillance devices with conventional light traps and BG-traps at the same locations. Their conclusion was straightforward: AI can help modernize vector surveillance, but current systems still fall short on trapping efficacy and algorithm performance, so they shouldn’t replace conventional methods yet. (frontiersin.org)

That finding lands at a time when surveillance programs are under pressure to do more with faster, more granular data. AI-enabled traps promise automated counting, identification, and real-time transmission, which could reduce labor and improve responsiveness. But mosquito surveillance is only useful if the trap reliably samples the right insects in the field. CDC guidance continues to emphasize that different trap types serve different surveillance purposes: BG-Sentinel traps are often used for Aedes aegypti, Ae. albopictus, and Culex species, while CDC light traps capture a broader range of adult mosquito genera. (frontiersin.org)

In Zhejiang, the performance gap was notable. The study reported that BG-traps had the highest trapping efficacy, followed by light traps, while the intelligent devices captured substantially fewer mosquitoes overall. According to the paper, BG-trap efficacy was 5.75 times that of the light trap, 23.88 times that of one intelligent trap model, and 88.71 times that of the other. When researchers standardized the comparison to the same operating time windows, one intelligent trap performed similarly to the light trap, but still remained well below the BG-trap. The AI devices did, however, show some ability to reflect seasonal mosquito density trends, particularly one system that tracked directional changes more consistently than the other. (frontiersin.org)

The broader Zhejiang surveillance context helps explain why this matters. A separate 2024 province-wide integrated surveillance study, also published in Frontiers in Veterinary Science, monitored mosquitoes and other vectors across all 11 prefecture-level cities. That program found an average mosquito density of 16.03 mosquitoes per trap-night, with Culex tritaeniorhynchus and Culex pipiens pallens as the dominant species, and the highest densities in livestock sheds. Investigators also screened more than 27,000 mosquitoes for dengue, yellow fever, Japanese encephalitis, West Nile, Zika, and chikungunya viruses. All tested negative in that survey year, but the infrastructure shows how entomologic surveillance supports broader disease prevention and early warning. (frontiersin.org)

Direct outside commentary on this specific Zhejiang paper appears limited so far, but the wider literature is moving in the same direction as the authors’ conclusion. A recent review in Parasites & Vectors describes AI-based mosquito technologies as an emerging step in mosquito baiting and surveillance, while stressing the need for field validation. Other recent work on smart-trap systems has likewise focused on what these tools can add to routine surveillance, rather than treating them as plug-in replacements for established trapping methods. That’s broadly consistent with the Zhejiang team’s conclusion that the near-term future is hybrid, not fully automated. (link.springer.com)

Why it matters: For veterinary professionals, especially those working in population health, livestock systems, public health partnerships, or zoonotic disease preparedness, the study is a useful reminder that surveillance technology has to be judged on operational performance, not just digital features. Real-time dashboards and automated counts are appealing, but if a device undersamples vectors, it can distort local risk assessment. In farm-adjacent settings, where mosquito abundance can be high and vector-borne pathogens remain a concern, under-detection could delay control decisions or skew trend analysis. The practical takeaway is that AI may improve workflow and data handling, but conventional traps still set the benchmark for dependable field collection. (frontiersin.org)

There’s also a procurement angle. Programs considering AI-enabled surveillance hardware may need to think less in terms of replacement and more in terms of layered deployment: conventional traps for validated collection, paired with AI systems for continuous monitoring, faster reporting, or supplemental trend detection. That kind of hybrid design may be especially relevant in regions managing multiple vector species, where trap bias already complicates interpretation. (cdc.gov)

What to watch: The next phase will be whether manufacturers and public health agencies can improve both capture efficiency and species-recognition accuracy under field conditions, and whether future validation studies show AI systems performing reliably across habitats, seasons, and target species, especially Aedes. Until then, the evidence supports augmentation, not substitution. (frontiersin.org)