Acceleration of simulations by application of a kernel method in a high-resolution Lagrangian Particle Dispersion model.

The role of microscale atmospheric dispersion modelling is becoming increasingly important in air quality assessment, especially in residential areas, for regulatory purposes and to project pollution control strategies. Consequently, the use of these models in long-term studies and forecasting systems is growing. However, this modelling can be challenging because of the amount of time and CPU required for the simulations, especially if the computational domain has a significant extension. This article describes the application of a concentration calculation methodology to reduce the computational time needed by microscale Lagrangian Particle Dispersion Models (LPDMs). These models normally estimate the concentration with the box counting method: a 3D mesh is set, and the density is computed counting particles in each box. An alternative method based on the use of the statistical technique of kernel density estimation is proposed to determine the concentration from the particles’ position. The kernel method allows for a reduction of computational particles emitted during the simulation, guaranteeing a similar accuracy to that of box counting method. It enables therefore to optimize the overall simulation time and the required CPUs in order to improve time and cost, enhancing the efficiency of models and widening their application fields. The use of the kernel method to perform high-resolution simulations took the first steps with an application inside the LPDM of PMSS (Parallel-Micro-SWIFT-SPRAY) system to evaluate road traffic gases emissions in urban areas, enabling an 80% simulation time reduction. In this article, additional features of this method are developed within the Micro-SPRAY model and validated against two new test cases. The test cases consider microscale simulations of industrial sources emitting gases and particles, evaluated both inside a domain divided in tiles and inside a nested domains configuration. The existing kernel method is enhanced in order to estimate the pollutant concentrations of point sources and to compute also the corresponding deposition at building-resolving scale and in nested domains with different horizontal resolution.