Augmenting community-driven vector surveillance with automated image classification: lessons from the Artificial Intelligence Mosquito Alert (AIMA) system

Abstract

The Mosquito Alert (MA) platform leverages artificial intelligence to enhance community-driven mosquito surveillance by automatically identifying mosquito species from geolocated images submitted via a mobile app. This empowers the public to report both native and invasive mosquitoes of public health relevance, contributing to early detection and monitoring efforts. The Artificial Intelligence Mosquito Alert (AIMA) system integrates machine learning image classification within an automated backend pipeline to enable real-time triaging of submissions: critical reports are flagged for expert review, routine cases are classified automatically, and contributors receive immediate feedback fostering participant engagement. By automating routine identifications, the system reduces the burden on experts, allowing them to focus on complex or ambiguous cases that require taxonomic expertise. This study focuses on two AIMA operational periods in 2023 and 2024. We evaluate model updates and performance across these years, highlighting both progress achieved and remaining limitations under real-world citizen science conditions. The most reliably classified species across both models were Aedes albopictus and Culex sp., whereas Aedes aegypti remained difficult to identify. Despite its limitations, AIMA remains central to enabling scalable, responsive, and intelligent mosquito vector surveillance, substantially reducing the time experts must devote to routine identifications. Functioning as an Early Warning System (EWS), MA produces real-time distribution maps of invasive species and rapidly delivers actionable information to public health authorities, facilitating timely responses and intervention.

Work supported by funding from the European Union's Horizon Europe program: E4Warning (101086640); VEO (874735); AIM COST (CA17108).