High Quality Air Pollution Dispersion Modelling Using High Computational Performance Lagrangian Particle Model

High quality air pollution dispersion modelling over complex terrain using Lagrangian particle model consumes a lot of computational time. In this study has been determined that there are some air pollution situations where unnecessary computational time is spent like accumulation of air pollution in the domain during the calm or extra high emissions from sources due to temporary failure of exhaust filters. To reduce unnecessary spent computational time a solution based on the control of number of virtual particles has been found. Existing Lagrangian particle air pollution dispersion model has been upgraded into new high computational performance model using integration of three methods to control the number of particles in the Lagrangian simulations. First method based on artificial neural networks is used to directly control the number of emitted virtual particles by predicting meteorological situations when too much particles will start to accumulate in the domain. Second method represents the use of cell concentration estimation method based on kernel density to assure high quality of the results. And the third method is based on the clustering algorithm to decrease the number of virtual particles in special situations if the first method can not enough decrease the number of particles. This integration of three methods ensures approximately constant number of particles in the modelling domain during whole simulation and consequently also constant and high quality of modelling results. Enhanced air pollution dispersion model has also been validated on the Šaleška region data set which represents highly complex terrain located in Slovenia. Validation confirmed the good quality of modelling results where spent computational has been reduced up to 50%.