Air Quality Forecast

Particulate matter (PM2.5) is a common air pollutant that can cause health problems, especially for at-risk groups. Wildfires, industrial processes, and vehicle emissions are common sources of the small particles that make up PM2.5. These concentrations can vary widely over short distances, and across time. Our goal is to offer accurate, localized, and time-specific air quality predictions to complement existing programs. By forecasting PM2.5 concentrations, we can help people plan their activities and reduce their exposure to air pollution.

The Air Quality Forecast shows the predicted PM2.5 concentration for the next 12 hours. Also included are the PM2.5 concentrations for the past 48 hours (all times in UTC).

The EPA's AirNow program provides near-real-time air quality data and next-day forecasts for the United States and embassies around the world. Using these measurements from the AirNow program, we can provide near-term forecasts of PM2.5 concentrations, to augment the existing next-day AQI forecasts. For more information, visit airnow.gov.

The Air Quality Forecast values are generated using a geometric deep learning model, trained on seven years of historical data from the AirNow program. The model uses Edge Convolution and a novel geographic encoding scheme to predict PM2.5 concentrations at a range of scales from local to global. These scales are defined by the H3 hexagon grid system, which allows for efficient aggregation across scales.

How to use the map:

Use the zoom and pan controls to explore the map.
Use the model and time controls to explore the forecast.
Click the border, fill, and label buttons to toggle the map layers.