
Traffic & Weather Map Movie Forecasting
- Study high-resolution weather and traffic movies
- Predict weather map movies
- Predict traffic using traffic and weather map movies
Building on the success of the Traffic4cast 2019 and 2020 competitions, Traffic4cast+ is offering new challenges and opportunities. The competition adds 7 diverse cities encouraging participants to investigate hidden structural similarities in traffic patterns across cities.
The traffic data are augmented with additional weather channels with a high temporal resolution of 5-15 minutes and a spatial resolution of approximately 3-6 km. The weather data are provided in two forms: derived products, such as cloud coverage, precipitation, pressure, temperature and wind, and raw sensor data (12 spectral channels in the optical and near-to-thermal infrared range). The latter allows exploring additional valuable patterns recognized in raw signals.
The new challenges of the Traffic4cast+ are:
- weather prediction using weather map movies;
- traffic prediction using traffic map movies and weather map movies.


The competition data are derived from a large fleet of probe vehicles, live incidents feed, and static map data and are grouped into 100m x 100m x 5 min spatiotemporal cells. The dataset is augmented by a combination of dynamic features, encoding traffic volume, speed, direction, and incidents, and static features, describing the road junctions and points of interest, such as entertainment, food and drink venues, shopping, parking, etc.
The goal is to predict short-term large-scale traffic states in all the cities. Participants can investigate differences and similarities in traffic patterns between the cities, and explore master models trained on multiple cities. Participants are encouraged to explore the impacts of static features, as well as additional features, such as air pollution. The competition is chosen for the second year for the prestigious NeurIPS conference.
The incredibly creative solutions of the 2019 competition revealed surprising insights:
- Our simple movies provided sufficient information to make state-of-the-art traffic predictions.
- Additional traffic attributes (weather, road map properties) are of value for bonus challenges and improve more complex models.
- It may be possible to separate rules underlying traffic dynamics from city-specific forecasts (transfer learning).
The solutions presented at the Traffic4cast competition track at NeurIPS 2019 are published in the Proceedings of Machine Learning Research.

Data example from Traffic4cast 2019.