During RP2 (M19–M36), E4Warning shifted from building separate data streams to operating integrated pipelines that turn observations into calibrated risk outputs and mechanistic insight. Mosquito Alert scaled further and its AI triage (AIMA) was upgraded for faster, more robust near–real-time classification of heterogeneous citizen imagery. In parallel, IRIDEON’s VECTRACK matured via a consolidated IoT service/API, improved ML classifiers and automated device-health monitoring, with field validation in Spain, Greece and Brazil benchmarking AI outputs against expert/manual counts and documenting remaining limitations (context/temperature effects and occasional false positives).
These gains fed into operational modelling, notably the consolidated Barcelona workflow integrating AI-filtered citizen science, smart-trap streams, conventional traps and harmonised covariates with sampling-effort correction to deliver daily nowcasts and short-term forecasts for vector-control planning, complemented by a broader Spain-level structure for transfer. Host-related evidence also progressed through improved GPS/Sigfox biologging (enhanced antenna coverage enabling tracking of medium-sized passerines), large-scale extraction of dispersal traits from Movebank, and passive acoustic monitoring overlapped with professional distance-sampling transects to estimate abundance and quantify bias; WP7 added mechanistic contact information via blood-fed mosquito collection and molecular diet inference.
Environmental and ecological inputs expanded through an updated multi-source covariate backbone (EO, climate/reanalysis, demography and land-surface drivers), improved downscaling/forecast verification, and VIC-based hydrology variables (soil moisture, runoff, baseflow, evapotranspiration) extended toward European-scale production to test added value beyond precipitation/humidity proxies. For dengue, D-MOSS advanced toward a generalised framework across Vietnam, Sri Lanka and Malaysia, while “transmission fringe” emergence analyses and EU-facing importation modelling continued to strengthen evidence on where and when risk emerges. Overall, RP2 demonstrated that heterogeneous streams can be operationally integrated and validated to generate interpretable outputs at surveillance and control scales.
