BC Air Quality — PM2.5 Forecast

Machine-learning forecasts of fine-particulate pollution, 24 hours ahead, for British Columbia monitoring stations.

OpenAQ sensors + Open-Meteo weather → Snowflake / dbt features → PyTorch LSTM → precomputed hourly into Postgres

Forecast — 24h ahead

µg/m³

Model accuracy — last 7 days

µg/m³ MAE

no realized forecasts yet

Forecast for

model · refreshed by the batch pipeline

PM2.5 — observed vs model

Each prediction was made 24 hours in advance from the previous 48h of sensor + weather data. Amber = the next 24 hours.

Good (0–12)Moderate (12–35.4)Unhealthy for Sensitive Groups (35.4–55.4)Unhealthy (55.4+)

Stations

Colored by latest predicted AQI band — click a station to select it.

loading map…

Forecast accuracy (realized)

Past forecasts joined against what the sensors actually measured.

No realized forecasts for this station yet — rows appear once forecasts mature (24h after issue) and the eval job has run.