Abstract
The emergence of data analytics and AI is promising for tourism itinerary planning, taking into account time-varying environmental hazards, including fine-particulate exposure, heat stress, and storm-related outdoor risk, particularly in tropical cities. We present an ambient-aware itinerary optimizer that combines Copernicus Atmosphere Monitoring Service forecasts, Extreme Gradient Boost bias correction, conditional value-at-risk uncertainty handling, and a Constraint Programming–Satisfiability scheduler. The objective is to minimize PM2.5 exposure, wet-bulb globe temperature-based thermal dose, and rain- or extreme-event-related outdoor exposure while respecting daily visit limits and itinerary feasibility. We evaluate the framework in a two-year forecast replay in Vietnam for 2024–2025 using 200 stratified scenarios, 29 points of interest, two regional tourism clusters, and five baselines. All plans are evaluated under a matched per-day tourist-attraction visit budget, so dose reductions reflect scheduling decisions. Relative to a tourist guidebook baseline, the full system reduces thermal dose by 90.1% and PM2.5 dose by 31.2%. An air-quality-only baseline achieves similar PM2.5 exposure but incurs 1.97-fold higher thermal dose, showing the limitation of single-hazard planning.
Keywords:
air quality; thermal stress; tourism itinerary; CP-SAT; WBGT; PM2.5; forecast bias correction; XGBoost; Vietnam tourism