A Simulation-based Framework for Dynamic Light Pollution Prediction in Urban Air Mobility

Research Note

Urban air mobility (UAM) is a vision of city transport in which small electric aircraft that take off and land vertically, often called "flying cars," carry people and goods above the streets. Free from ground congestion, the concept is being developed and tested around the world, and in Japan demonstration flights at Expo 2025 in Osaka marked a step toward real deployment.

When many such vehicles fly over a city, one factor that may become a problem alongside noise and safety is light. The surface of an aircraft reflects sunlight, and when the angles line up, the reflection reaches the eyes of people on the ground as an intense glare. The phrase light pollution usually brings to mind artificial lighting brightening the night sky, whereas this is a daytime phenomenon caused by moving vehicles, a dynamic form of light pollution. Because both the aircraft and the sun keep moving, where and when the glare occurs changes from moment to moment.

Such glare poses a real risk to drivers and pedestrians, so it needs to be predicted and assessed across the whole city at the stage where flight routes and no-fly zones are designed. That computation is far from easy, since physically faithful simulation of the reflection is too slow to cover an entire city, while replacing it with machine learning struggles to generalize across complex building geometries.

This work proposes a hybrid framework that combines the two. A neural surrogate trained in an unoccluded environment approximates the dominant reflection from the aircraft, and the effect of occlusion by buildings is incorporated as a lightweight geometric attenuation. Evaluated on digital twins of three Japanese cities, the method keeps accuracy (reducing RMSE by 5 to 30% with an F1-score above 0.8) while cutting the evaluation of a 5-second prediction horizon for 10,000 observation points from more than 3 hours of physical simulation to 1.62 seconds, fast enough for city-scale applications such as dynamic light pollution risk maps and adaptive no-fly zone design.