Optimization of Plume Model Calculations and Measurement Network with a Kalman Filter Approach

Abstract

In many industrial regions there is a strong demand for accurate monitoring of the air pollution and its sources. The Rijnmond area around Rotterdam in the Netherlands is an example of an industrial area affected by air pollution through many industrial and traffic activities as well as shipping emissions. In the area the air quality is traditionally modelled based on a Gaussian plume model using local emissions. To estimate the background concentration due to transport from non-local sources, the average difference between the model calculations and observations at three stations is taken. However, in case of local high emission events, this difference cannot be pointed to the background and the simple approach leads to false estimates of the background resulting in over or under estimation of the concentrations in the rest of the area. In this study we have developed a modeling system with a Kalman filter approach to optimize plume model concentrations using actual observations. This system is capable to adapt modelled concentrations based on the originating source of the concentrations, more accurately than using simple background estimates. We will present the system set-up and results for a testcase in the Rijnmond area for NO_x. For this testcase we have predefined the ‘normal’ concentrations for different meteorological situations with a Gaussian plume model. Those model calculations are put in a Kalman filter system and assimilated with actual observations. In case of a measured difference of concentration compared to the model, the system will adapt the most likely sources and in addition provide an uncertainty range of the calculation. The results show the system is much better able to represent the NO_x concentrations than previous system. Finally we will show how the system can be used to optimise the monitoring network through minimization of the uncertainty of the model results.

Questions and Answers

Questioner: Heinke Schluenzen

Question: How did you consider the spatial dependency of emissions (e.g. from ships), which are due to different positions and movements of the ships time dependent? How is this considered when combining the pre-calculated fields?

Answer: At this point, we take an annual average of the emissions, so we do not take into account the time-dependency due to shipping movements. Contrary to road traffic we have a flat temporal pattern for shipping emissions, so we do not have any timing for shipping, such that only the Kalman filter introduces the some time pattern for shipping.

Questioner: Georgios Tsegas

Question: Regarding the shipping emissions, are they composed from isolated elevated point sources?

Answer: The emissions are not constructed from isolated points, but from main shipping tracks and large harbours. According to those shipping tracks and the staying time and movements in the harbour, the emissions are calculated as an annual average.

Questioner: Luca Delle Monache

Question: What is the forecast lead time of the prediction system you have described? Is there a positive impact of the Kalman filter step beyond 1 h.

Answer: The forecast step is one hour because the measurements are on a hourly basis. After each hour the Kalman filter is applied and the current ‘state’ is improved. With this improved current state, we make a better forecast for the next hour, which will be again updated after an hour. So in general the impact of the Kalman filter calculation on timestep ‘k’ is very limited after time k + 1.