Models / hbku

hbku/air-quality-forecaster

Air Quality Forecaster

PublicMinimal riskOpenForecastingTensorFlowN/A1.2.0
1.9K 15K/30d 95

MAE PM2.5 (µg/m³, 24h)

6.8

MAE PM10 (µg/m³, 24h)

18.4

AQI-band accuracy (72h)

0.81

Dust-event recall

0.79

About this model

Produces 72-hour PM2.5 and PM10 forecasts for each national air-quality monitoring station, blending station history with meteorological drivers including shamal wind events. Outputs hourly concentrations plus AQI-band probabilities for public-advisory thresholds.

Intended use

Public health advisories, outdoor-event planning, and school activity guidance driven by station-level dust and particulate forecasts.

Training data lineage

Trained on the air-quality-stations archive of hourly measurements from 2020 through 2025 with ERA5-derived meteorological covariates.

Limitations & bias notes

Rapid-onset shamal dust fronts are captured with a lag of two to four hours, under-forecasting the initial spike. Stations near active construction corridors show localised bias not explained by the meteorological features.

Evaluation metrics

MAE PM2.5 (µg/m³, 24h)6.8
MAE PM10 (µg/m³, 24h)18.4
AQI-band accuracy (72h)0.81
Dust-event recall0.79
#air-quality#pm25#forecasting#environment#dust-storms#public-health

Try it

Live sandbox

horizon=12

fixture

Feedback

Owning entity

hbHBKU
Updated
2026-04-30
Latest version
1.2.0
License
Open
Access
Open