Virtual Height Dispersion Model: evaluation against tracer data in different turbulent regimes.

Transcription

1 Virtual Height Dispersion Model: evaluation against tracer data in different turbulent regimes. (a) ISIAt A/CNR Strada Provinciate per Arnesano, Lecce Abstract The Virtual Height Dispersion Model is an air quality dispersion model developed to evaluate ground level concentrations from tall stacks ( Rizza et al. ). It relies on a new gaussian formulation proposed by Lupini & Tirabassi^ for transport and diffusion in the Atmospheric Boundary Layer (ABL). In that formulation, the real source height is replaced by a virtual height related through wind and eddy diffusivity profiles to stability conditions of ABL. A performance evaluation of the model to different turbulent regimes was performed with tracer data from some experimental field campaigns. 1 Introduction The VHDM (Virtual Height Dispersion Model) is a new generation gaussian type model for evaluating the ground level concentration of nonreactive gases released from tall stacks. The model describes dispersion processes in terms of basic scaling parameters, such as the height above ground, the height of boundary layer, the Monin- Obukhov length scale, the friction velocity and convective velocity, getting over the traditional discrete stability classes. The mathematical basis of the model together its preliminary evaluation has already been described in a precedent paper ( Tirabassi & Rizza*), so that, here, only a short description of its main features will be given. The aim of this work is to evaluate general performances of VHDM for predicting atmospheric dispersion in different turbulent regimes during unstable conditions occurring in several experimental situations. To this purpose we tested the model on tracer data from non buoyant tracer

2 346 Air Pollution Modelling, Monitoring and Management experiment in residential area (Copenhagen data set), as well as on plume measurements from buoyant power-plant in urban area (Indianapolis data set). 2 Model description The model is based on the well-known gaussian plume formulation. The ground level concentration is given by: where Cy is the cross-wind integrated concentration, and Gy is the horizontal dispersion parameter. The integrated crosswind concentration is evaluated by a Fickian-type formulation, in which the real source is replaced by a virtual source height function of the vertical profiles of wind and eddy diffusivity (Lupini & TirabassP): i&_ (1) where Q the source strength, Us and Kg are the wind speed and the eddy diffusivity at the source height respectively, 1, and ^ are two 'virtual' source heights defined as follows where ZQ is the roughness length, Hg is the effective release height, u(z) and KZ(Z) are the wind velocity and the eddy diffusivity vertical profiles respectively. The model accepts both experimental and theoretical profiles for eddy diffusivity and wind velocity, provided the integrals in equations (2) exist. Plume rise and eventual plume penetration of elevated stable layers are modelled with the formulation suggested by Briggs\ 2.1 Wind speed The wind profile is assumed to vary only in speed as a function of height. The dependence of wind speed on height is computed using the well known similarity functions: 7 / (3) ^mo J \ ^mo where u, is the friction velocity, k is the von Karman constant, L^o is the Monin-Obukhov length and \f^ is a universal similarity function computed using the software library from Beljaars et ala

3 Air Pollution Modelling, Monitoring and Management 347 Equation (3) is only valid in the Atmospheric Surface Layer (SL) defined as the bottom 10% of the ABL, in the upper part (up to mixing layer height H) the wind is almost constant with height, hence: u(z) is computed by (3) for z< zy u(z) =u(zb) for z> Zb where zy = max( 0.1 H, Lmol). 2.2 Vertical dispersion Turbulence within the ABL is described using a scaling approach (Holtslag & Nieuwstadt^). Following this scheme the ABL is subdivided into various regions, each governed by different length and velocity scales. The various turbulent regimes are characterised by two dimensionless parameter H/Lmo and z/h (for further details and nomenclature see the cited paper). In the SL, turbulence is mostly caused by eddy shear stress, and eddy diffusivity profile can be derived directly from the flux profile relations of Monin Obukhov theory: The non-dimensional concentration profile function <f>c is assumed to be similar to that of heat 0& Businger et af. Above the SL, during stable and near neutral conditions, (i.e. for H/Lmo>-10) the SL formulation is extended by an empirical function of z/h as recommended by Troen & Mahif K =, In the mixed layer portion of the convective boundary layer(i.e. for H/Lmo<-10), buoyant production of turbulent kinetic energy is much more important than shear production. Therefore, friction velocity is replaced by the convective velocity w* as the scaling velocity in order to give, for the vertical eddy diffusivity, the following relation (Wyngaard & Brost* ): (6) 2.3 Horizontal dispersion The lateral dispersion parameter is computed following the model proposed by Berkowictz et al.^. This is based on the assumption that the turbulent intensities and related timescales can be expressed as a sum of two separate terms, the first is the contribution from convective turbulence the other one from mechanical turbulence, that is: O.y = (C ^com, + O ^meca )^ X / W^ (7)

4 348 Air Pollution Modelling, Monitoring and Management where Umed is the wind speed averaged over the layer between the ground and the source height. The contribution from convective turbulence is given by: G"c = (0.25^ / ( ^ / (#%^ )) while that one from mechanical turbulence is : G^mech =(%*) taking in account the internal turbulence induced by the buoyancy in thermal plumes the effective dispersion parameter becomes: where DH is the plume rise. 3 Field experiments The field data utilised to test the model were selected from two field experiments on atmospheric dispersion according to the protocol agreed at the Manno Workshop (Cuvelier^.) 1. The Copenhagen data set (Gryning^) involved an elevated non buoyant SF6 release in a mainly residential area. The surface sampling arrays were positioned in arcs from 2 to 6 km downwind of the source. 2. The Indianapolis SF^ data base (TRC^) involved an elevated buoyant release in an urban area with flat terrain. There were 177 ground monitors in arcs from 0.25 to 12 km downwind of stack. Only data with a quality factor 3 were considered. The main characteristics of the two datasets are listed in Table 1. Table 1: Characteristics of the two sites used for model evaluation. Site Name Copenhagen Indianapolis Source 115 m non-buoyant 83 m buoyant Number of monitors (distance from source) about 60 (2-6 km) about 160 ( km) Surrounding terrain flat-mainly residential flat, urban area 4 Model results and discussion Following the scaling approach from Holtslag & Nieuwstadt^, in the unstable boundary layer (L^o < 0) five distinct regimes are identified, depending on scaling parameters H/Lmo and z/h: the Surface Layer (SL) for z/h < 0.1 and z/lmo < 1, the Free Convection Layer (FCL) for z/h <0.1 and z/l > 1, the Mixed Layer (ML) for 0.1< z/h < 0.8 and H/Lmo > -10, the Near Neutral

5 Air Pollution Modelling, Monitoring and Management 349 Upper Layer (NNUL) for 0.1< z/h < 0.8 and H/L < -10 and the entrapment layer (EL) for z/h>0.8. The model performances were tested in these distinct regimes for each data base, so that results obtained are classified by stability using mainly 1) scatter plots, 2) residual plots for arcwise maximum concentration, 3) Quantile- Quantile plots. In the first two cases the data are paired in time and distance, whereas in the last case the same data are unpaired in time or space. The residual (observed/predicted concentrations) plots as function of input model parameter are useful tools to identify systematic problems with model formulation (signified by a trend in the residual plot). A model's performance should not show a trend with input or calculated parameters, but only some scatter about a residual of 1.0. The Q-Q plots are simple pairings of modelled concentrations, ranked from highest to lowest, with measured concentrations ranked in the same manner. They are useful for comparing the distribution of the two data sets. The performances measures are obtained using the statistical evaluation procedure software described by Hanna^ and are defined in the following way : cor (correlation^ (C,-Q)(C,-C^/o,o, fa2 = fraction of Co values within a factor two of corresponding Cp values fb (fractional bias)= where subscripts o and p refer to observed and predicted quantities, and an overbar indicates an average. 3.1 Passive release The Copenhagen experiment comprises only nine runs, covering most of the unstable boundary layer turbulent regimes. Therefore, we didn't limit to analyse the maximum arcwise concentrations but we investigated the complete ground pattern and we compared the VHDM predictions with data measurements at each receptor along the arcs. Figure 1 shows the scatter plot between model results and concentration data, whereas Table 2 summarises the statistical indexes. It is clear the correlation between the data in each regime considered; best agreement is found for the SL one, while a bigger scatter is found for more convective cases. In each regime, anyway, there are some discrepancies in the region of lower values corresponding to concentrations measured along the arc edges. This can be due both to the increased uncertainty in measurements with low concentration and to the lateral dispersion parameterisation utilised, the last one should be further investigated by using other data sets.

6 350 Air Pollution Modelling, Monitoring and Management 1800-, " " observed observed 1800 Figure 1: Copenhagen data set. Comparison of predicted and observed ground level concentration in different turbulent regimes. Data are normalised to the emission rate. Table 2: Copenhagen data set. Summary of performance measures. SL NNUL FCL ML Cor fa fb In Figs.2a and 2b the analysis is moved on the arc maximum concentrations. There are too few data to extrapolate general consideration, anyway, most of the data are in factor two with a tendency for the model to underpredict in the very convective cases.

7 10 Air Pollution Modelling, Monitoring and Management if ^% Jdoo distance (m) observed 4000 Figure 2: Copenhagen data set a) residual plot for the maximum concentration on all arcs and all experiments, b) Q-Q plot of predicted versus observed arcwise maximum concentrations. The squared are for SL regime, the circles for FCL, the stars for NNUL, the triangles for ML regime. Data are normalised to emission rate. 3.2 Buoyant release The Indianapolis experiment involved a very buoyant plume. From data base we selected only experiments occurred during unstable conditions. Because of buoyancy, the effective source height was always over the SL and the FCL, so that only ML and NNUL regimes are present. Here, the evaluation is focused on maxima concentrations, where the maximum of the observed concentration on an arc of monitoring stations is interpreted as being the centreline concentration. The comparison of VHDM estimates to the Indianpolis data is found in figs. 3a and 3b, while in Table 3 the statistical indices are given. There is an increasing scatter in data respect to Copenhagen experiment mainly at shorter distances and for lower values. This can be explained by 1) some simplification in the model while treating the buoyancy and the urban effects that affect the plume dispersion 2) a larger uncertainty in the measurements. The residual plot, anyway, shows a satisfactory correspondence between the distribution of model estimates and observations mainly in the upper part of distributions. 4 Conclusions The VHDM is an operational advanced model to be used in regulatory air pollution applications, particularly when emission derives from industrial stacks. The model performances have been evaluated using two data sets involving different experimental situations. Overall comparison between the

8 352 Air Pollution Modelling, Monitoring and Management model results and the measurements shows that the model reproduces in a realistic way the ground level concentration pattern in each conditions. Due to the model simplicity the scatter between data and model results becomes larger with growing complexity of experimental situations, anyway the model can predict properly the frequency distribution of concentrations, or merely the highest concentrations. o O ,1-0,01- m 0, distance (m) ^ 100 CD fie- 0,1 0, observed Figure 3. Indianapolis data set a) residual plot for the maximum concentration on all arcs and all experiments, b) Q-Q plot of predicted versus observed arcwise maximum concentrations. The stars for NNUL, the triangles for ML regime. Data are normalised to emission rate. Table 3: Indianapolis data set. Summary of performance measures NNUL ML Cor fa fb References 1. Rizza U., Mangia C. & Tirabassi T. Validation of an operational advanced Gaussian model with Copenhagen and Kincaid datasets, Int. J. of Environment and Pollution (in press). 2. Lupini R. & Tirabassi T A simple analytic approximation of ground-level concentration for elevated line sources. J. Appl. Meteor.,1981, 20, Tirabassi T. & Rizza U. Applied dispersion modelling for ground-level concentrations from elevated sources. Atmos. Environ., 1994, 28,

9 Air Pollution Modelling, Monitoring and Management Briggs G.A. Plume rise predictions, Lectures on Air Pollution and Environmental Impact Analyses, ed D.A. Haugen pp , American Meteorological Society, Boston MA, Beljaars A.C.M. & Holtslag A.A.M. A software library for he calculation of surface fluxes over land and sea, Environ. Soft., 1990, 5, Holstlag A.A.M. & Nieuwstadt Scaling the atmospheric boundary layer. Bound. Lay.Met.,1986, 36, Businger J.A., Wyngaard J.C., Izumi Y. & Bradley E.F. Flux-Profile Relationships in the Atmospheric Surface Layer, J. Atmos. Sci, 1971, Troen IB. & Mahrt L. A simple model of the atmospheric boundary layer; sensitivity to surface evaporation, Boundary-Layer Meteorol., 1986, 27, Wyngaard J.C. & Brost R.A. Top-down and bottom-up diffusion of a scalar in the convective boundary layer. J. atmos. Sci.,1984,41, lo.berkowicz R.R., Olesen H.R. & Torp U.. The Danish Gaussian air pollution model (OML): description, test and sensivity analysis in view of regulatory applications. Proc. NATO-CCMS 16th Int. Meeting on Air Poll. Modelling and Its Applications, C. De Wispelaere, F.A. Schiermeier and N.V: Gillani Ed. (Plenum Press, New York, N.Y. (USA), pp , ll.cuvelier, C. (Editor): Workshop on intercomparison of advanced practical short-range atmospheric dispersion models. Manno 1993, Joint Research Centre (European Commission Institute for Safety Technology), EUR EN, Gryning, S.E. Elevated source SF&-tracer dispersion experiments in the Copenhagen area, Report R-446. Risoe National Laboratory, Roskilde Denmark, TRC, Urban power plant plume studies EPRI Report EA-5468, EPRI 3412 Hillview ave., Palo Alto CA 94303, Hanna, S.R Confidence limits for air quality models, as etimated by bootstrap and jackknife resampling methods, Atmos. Environ. 1989, 23,