8 th Int. Conf. on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes ______________________________________________________________________________________ - 19 - VALIDATION OF THE LAGRANGIAN PARTICLE MODEL DIPCOT FOR MESOSCALE AND LONG-RANGE ATMOSPHERIC DISPERSION OVER COMPLEX TERRAIN E. Davakis 1 , S. Andronopoulos 1 , J.G. Bartzis 2 and S. Nychas 3 1 Environmental Research Laboratory, Institute of Nuclear Technology and Radiation Protection, NCSR “Demokritos”, GREECE 2 Department of Energy Resources Management Engineering, Aristotelian University of Thessaloniki, GREECE 3 Department of Chemical Engineering, Aristotelian University of Thessaloniki, GREECE INTRODUCTION The stochastic (Lagrangian) particle atmospheric dispersion model DIPCOT (Davakis et. al, 2001, Davakis et. al, 2000) is evaluated by simulating a mesoscale (TRANALP campaign) and a long-range (European Tracer Experiment-ETEX) dispersion experiments. The effect of the method of calculation of the pollutant concentration is also examined, by comparing the results of three methods: a box counting method with variable box dimensions and two Gaussian- shaped density kernels. The evaluation procedure is based on the statistical and graphical comparison of the predicted concentrations against the observed ones, using some well-known performance indices and various types of plots. THE DISPERSION MODEL AND THE CONCENTRATION CALCULATION The Lagrangian particle dispersion model DIPCOT is a 3D model aiming at the study of atmospheric dispersion over complex topography. The mass of a pollutant is distributed to a certain number of fictitious particles, which are displaced within the computational domain, following the wind flow, and assuming that turbulent diffusion can be modelled as a Markov process. Calculating the trajectories of these particles simulates atmospheric dispersion. Further details about DIPCOT can be found in Davakis, S. et. al. (2000). In the Lagrangian atmospheric dispersion models two methods are used for the estimation of concentration at a given location: the box counting method and the density kernel estimation method. In the box counting method the concentration is computed by counting the number of the particles that fall into an imaginary volume around the location of interest. There are no strict rules determining the size of the sampling box. However, a too small volume would lead to large fluctuations of concentration, while a too large volume, would result in over-smoothed concentrations. This problem can be overcome (de Haan, P., 1999, Uliasz, M., 1994) by increasing the number of released particles, in expense, however, of computing time and resources. In the density kernel method the mass of a particle has a specified spatial distribution, called density kernel (de Haan, P., 1999). The concentration at the location of interest is calculated by summing the contributions of all the particles. The number of particles that can be used in this method is smaller than in the box counting methods (e.g., Yamada, T. and S. Bunker, 1988). The main parameters that must be predefined are the shape of kernel and its bandwidth (i.e. the width as a function of the particle mass distribution). In this paper the following concentration calculation methods are intercompared in the framework of the model DIPCOT: i) a box counting method (from now on called as CM1) as described by (Uliasz, M., 1994), with a modification concerning the box dimensions that are set equal to the maximum variance of the displacement of the all particles at each dimension, ii) a Gaussian-shaped density kernel method (from now on referred to as CM2), where the bandwidths are equal to the variance of the particle dispersion σ x , σ y , σ z , as proposed by Yamada,