Kalman filter in air pollution modelling
- The text on this page is taken from an equivalent page of the IEHIAS-project.
An air pollution model (URBIS) was coupled with a monitoring network to improve estimates of exposure to air pollutants in the Rotterdam area. The air pollution model was very detailed and consisted of plume models to calculate dispersion from shipping, road traffic and large point sources such power plants and industry. Exposure to road traffic in street was estimated using the semi empirical street model CAR. The model uses detailed emission inventories and employs a 100 m resolution. Outputs of the model are detailed maps of concentrations.
This version of the model (URBIS-Real time) uses results of earlier calculations in a look up tables to calculate concentrations every hour. The actual wind speed and direction are used to interpolate between the results of earlier calculations. The calculated concentrations are then linked to the measurements to improve the estimates. Two methods were compared: in the first method the results of measurements on three stations are used to calculate the contribution of the background to the concentration. In the second methods a more complex Kalman filtering technique was employed. Every hour the results of calculated concentrations are compared with measured concentrations on eleven stations in the area. On the basis of each comparison an improved emission inventory is made and the improved inventory is used in the next hour.
Comparison of the two methods showed that the Kalman-filter is a serious improvement compared to the simple background corrections. The model treated with the Kalman filter leads to a reduction of the uncertainty of 27 % compared to the non treated model. The improvement was very strong in areas with no monitoring.
Explanation of the method
Details of the methodology are provided in the attached report.
