You are on page 1of 8

Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries


journal homepage: www.elsevier.com/locate/jlp

Computational uid dynamics simulation of fog clouds due to ambient air vaporizers
Filippo Gavelli*
Exponent, Inc., 17000 Science Dr., Suite 200, Bowie, MD 20715, USA

a r t i c l e i n f o
Article history: Received 23 February 2010 Received in revised form 19 July 2010 Accepted 30 July 2010 Keywords: CFD Fog Atmospheric dispersion Ambient air vaporizer LNG

a b s t r a c t
Ambient air vaporizers (AAVs) are widely used to regasify liqueed industrial gases, which are liqueed for transport and storage. Depending on the conditions (temperature and relative humidity) of ambient air and AAV efuent, the potential exists for the formation of fog as the two uids mix with each other. This has raised some regulatory and environmental concerns that the fog cloud may impact human activities in the vicinity of the AAV arrays. This paper describes a CFD-based modeling approach to predict the formation, dispersion and dissipation of a fog cloud due to AAV operation. The model uses the psychrometrics equations to determine when saturated air conditions are reached and to calculate mass and energy transfer between the moist air and the fog cloud. A parametric study is presented, based on an array of 6 AAVs, to demonstrate the effects of wind speed and AAV discharge elevation on the behavior of the fog cloud. 2010 Published by Elsevier Ltd.

1. Background Ambient air vaporizers (AAVs) are widely used to regasify liqueed industrial gases, which are liqueed for transport and storage. Examples of AAV use include the regasication of liquid oxygen in hospitals and liquid nitrogen or argon in metal working facilities. More recently, large scale applications of AAVs have been proposed for liqueed natural gas (LNG) import terminals. AAVs use ambient air as the source of energy to regasify the cryogenic liquid. Air ows by natural or forced draft on the outside of long, vertical nned tubes and transfers heat to the cryogenic liquid within the tube (see Fig. 1). The use of heat from ambient air results in lower operating costs and lower environmental impact than other regasication methods based on fossil fuel combustion or open loop water circulation. Nonetheless, regulatory and environmental concerns have been raised about the potential impact of AAVs used in LNG import terminals, due to the large number of vaporizers involved in those applications (approximately one hundred or more units, covering a footprint on the order of a thousand square meters or more). Of particular interest is the formation of fog as the cold air discharge from these large AAV arrays mixes with warm and humid air, and the potential for the fog

cloud to affect human activities (e.g., transportation) in proximity of the LNG terminals. This paper will review the physics of fog formation and will present a method to quantify, using computational uid dynamics (CFD), the formation and dispersion of fog from the operation of an array of AAVs. 2. Principle of operation of AAVs Ambient air vaporizers are long, single-pass heat exchangers with the cold uid (the cryogenic liquid) owing inside vertical tubes and the warm uid (ambient air) owing on the outside (see Fig. 1). The AAV tubes are nned on the outside to increase the heat transfer surface on the air side. The tubes are aligned vertically to take advantage of the buoyancy-driven downward air ow that is established as ambient air cools down between the nned tubes. In order to increase the buoyancy-driven ow, as well as the heat transfer surface, AAV towers are generally tall and slender (approximately 3 m by 3 m footprint and 12 m tall, in many LNG applications). The discharge end (at the bottom of the towers) is elevated above ground to reduce the overall ow resistance and to allow the cold discharge (the efuent) to be dispersed (see Fig. 2). The energy extracted from the air stream consists of both sensible and latent heat losses. The sensible heat loss results in cooling of the air stream. The latent heat loss is associated with the condensation of part of the water vapor (moisture) in the air, once the air has been cooled down below its dew point temperature. The condensed moisture is deposited on the nned walls of the vaporizer tubes. If the

* Current address: GexCon US, Inc., 7735 Old Georgetown Rd., Suite 1010, Bethesda, MD 20814, USA. Tel.: 1 301 915 9925. E-mail address: fgavelli@gexcon.com. 0950-4230/$ e see front matter 2010 Published by Elsevier Ltd. doi:10.1016/j.jlp.2010.07.009

774

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

as the efuent temperature is by design well below the dew point of ambient air. As the efuent mixes with ambient air, which is warmer and has higher moisture content, both the temperature and the moisture content of the mixture change. Under the assumption of negligible heat transfer to/from the environment (i.e., adiabatic mixing), the mixing of cold air efuent and ambient air can be traced graphically on a psychrometric chart (see Fig. 4). The psychrometric chart allows the properties of the mixed state (e.g., enthalpy, humidity ratio, etc.) to be determined by knowing the conditions of the mixing streams (i.e., ambient air and efuent) as well as their respective mass fractions, as follows (McQuiston, Parker, & Spitler, 2000):

um xA uA xE uE
hm xA hA xE hE
where:      

(1) (2)

A denotes the ambient air; E denotes the efuent air; m denotes the mixed air; u is the humidity ratio (mass of moisture per mass of dry air); h is the enthalpy (kJ/kg). x is the mass fraction of the uid (efuent or ambient air) in the mixture.

For any given temperature and pressure, the maximum amount of moisture that can be in the vapor phase is given by the saturation humidity ratio:
Fig. 1. Ambient air vaporizer with nned tubes gasifying liquid natural gas.

nned wall temperature is below freezing, the condensed moisture will freeze on the walls, forming a soft, snow-like frost blanket that grows during operation from the base of the AAV towards the top (see Fig. 3). The frost blanket creates thermal insulation between the cryogen and the air, and therefore decreases the heat transfer efciency of the vaporizer. When the cryogen ow is interrupted, the nned tube walls warm up and the ice falls off, returning the AAV to its original heat transfer efciency. Different types of ambient air vaporizers are available. A distinction exists between natural draft and forced-draft AAVs. Natural draft AAVs rely solely upon buoyancy to drive ambient air ow through the units. As such, they have no moving parts and result in the lowest operating cost. However, their performance is strongly dependent upon ambient air conditions and very sensitive to ow obstructions created by ice formation. Forced-draft AAVs instead utilize a set of fans to supplement the buoyancy-driven ow of ambient air through the unit. Therefore, forced-draft AAVs typically have higher air ow rate (i.e., higher regasication capacity per unit) and are less sensitive to ambient air conditions. However, operating costs are increased by the presence of the fans. Another distinction exists between direct and indirect vaporizers. Direct vaporizers transfer the heat extracted from ambient air directly to the cryogenic liquid, which is owing inside the AAV tubes. Indirect vaporizers, instead, utilize an intermediate uid that ows through the AAV tubes to absorb heat from ambient air and then transfers it to the cryogenic liquid in a separate heat exchanger. For a discussion of the pros and cons of either system, the reader is directed to Shah, Wong, and Minton (2008). 3. Fog formation The efuent discharged from an ambient air vaporizer is typically at temperatures near or below freezing. It is also at saturation,

us 0:622

Ps Pa Ps

(3)

where Pa is the local pressure and Ps is the vapor pressure of water vapor at saturation. Ps can be calculated from the MagnuseTetens formula (Barenbrug, 1974) as a function of local temperature (in degrees Celsius):

  17:27T kPa Ps 0:6105 exp 237:7 T

(4)

If the mixture humidity ratio is greater than us, the mixture is supersaturated and the excess moisture (um us) will condense into water droplets e commonly known as fog (see point M in Fig. 4). The fog concentration F will then be given by:

F rm um us
where:

kg m3

 (5)

 rm is the density of the mixed air stream (kg/m3);  us is the saturation humidity ratio at the saturation temperature for the enthalpy hm of the mixed stream. Whether fog is formed by the mixing of efuent and ambient air depends on the physical states (e.g., temperature and relative humidity) of the mixing masses. In general, fog will be more likely to form in higher ambient air temperature and relative humidity. The effect of ambient moisture is shown in Fig. 5: if the efuent (0  C, 100% RH) mixes with ambient air at 30  C, 80% RH fog is formed (red line), while if the same efuent mixes with air at 30  C, 40% RH fog is not formed (blue line). For a given efuent state, the range of ambient conditions that will generate fog can be identied graphically as those ambient conditions that fall in the shaded area above the line passing through the efuent state and tangent to the

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

775

Fig. 2. Array of natural draft ambient air vaporizers.

saturation line. The lower boundary of the fog region is graphically shown as the green line in Fig. 5. Fig. 6 shows the variation of relative humidity as a function of the degree of mixing (DoM) of the efuent and ambient air in the mixed air stream, for the same two cases shown in Fig. 5. The DoM in Fig. 6 ranges from 0 (pure efuent) to 1 (pure ambient air), so the evolution of a mass of efuent as it moves away from the AAV and mixes with ambient air can be visualized by moving from left to right on the plot. The red curve shows that RH > 100% (i.e., fog forms) as soon as mixing occurs, and fog persists until the DoM > 0.74, that is, until the efuent is diluted to a ratio of approximately 1e2.8 with ambient air. The blue curve in Fig. 6, instead, never crosses into the fog region (i.e., the relative humidity always remains below saturation or RH < 100%).

A criterion to determine when an atmospheric release should be treated as a dense gas release is provided by Britter (1989) in terms of a modied Richardson number. According to Britters formulation, an atmospheric release behaves as a dense gas release when:

 rr a g

ra

Q0 D

1 3 ! 0:15 (6)

U
where:      

g is the acceleration due to gravity [m/s2]; r is the density of the efuent [kg/m3]; ra is the density of ambient air [kg/m3]; Q0 is the total efuent volumetric ow rate [m3/s]; D is the characteristic dimension of the source [m]; U is the ambient air velocity [m/s].

4. Fog dispersion modeling The cold air efuent from ambient air vaporizers for LNG regasication is approximately 25e45  C colder than the ambient air. Therefore, the density of the efuent is approximately 15e35% higher than the density of ambient air. Before the dispersion of a fog cloud from AAV operation can be modeled, it is important to understand whether the fog cloud behaves as a passive gas (i.e., it is advected by the existing air ow distribution) or a dense gas (i.e., it straties near the ground and affects the pre-existing air ow patterns). The modied Richardson number is found to be greater than 0.15 for most LNG-specic applications of AAVs. Therefore, the behavior of a fog cloud from an array of AAVs can be expected to be similar to that of other dense gas clouds (e.g., LNG vapor clouds): the fog cloud will tend to remain close to the ground, and mixing with ambient air will occur primarily by shear-induced turbulence at the interface between the two uids. This paper presents a method to perform quantitative analyses of AAV-induced fog dispersion using commercially available CFD models. The work presented in this paper was performed using

776

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

Fig. 4. Adiabatic mixing of efuent (E: 0  C, 100% RH) and ambient air (A: 30  C, 80% RH). Mixed state (M) corresponds to 50% dilution of the efuent.

Fig. 3. Active AAV with frost blanket due to condensed moisture (left) and inactive AAV (right).

Star-CCM, version 4.02 (www.cd-adapco.com); however, a similar method may be implemented with other CFD models. The overall approach to a fog dispersion simulation using CFD is similar to that followed for other dense gas releases (Gavelli, Bullister, & Kytomaa, 2008; Luketa-Hanlin, Koopman, & Ermak, 2007):  The simulation domain must be dened, including all geometrical features that will affect the dispersion of the fog cloud. In this case, the presence of the AAV towers must be accounted for, as it is likely to have a strong effect on the ow eld and turbulent mixing near the efuent discharge;  The simulation domain needs to be discretized using a mesh that can accurately resolve the areas where fog is present. This is particularly important near the ground, where the stratication induced by the dense cloud typically requires a high grid resolution in the vertical direction. A grid independence study or a Richardson extrapolation (Carey, 1997) should be performed to ensure that the results are not affected by the meshing choices;  Initial and boundary conditions must be imposed for velocity, temperature, and turbulence, as well as, for the ambient concentration of air and water vapor. Boundary conditions consistent with the PasquilleGifford stability classes (Turner, 1964) should be imposed, unless sufcient data is available from local weather stations.

Fig. 5. Mixing of efuent (0  C, 100% RH) with ambient air. The red line (with ambient air at 30  C, 80% RH) crosses the fog region; the blue line (with ambient air at 30  C, 40% RH) does not. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780 Table 1 Fog dispersion distance for the different scenarios. Scenario Wind speed (m/s) 2 m/s 2 m/s 2 m/s 5 m/s 5 m/s PasquilleGifford class F F F D D Discharge elevation 1.5 m 3.0 m 6.0 m 1.5 m 3.0 m

777

Max. fog dispersion distance 58 m 55 m 34 m 44 m 41 m

X2F05 X2F10 X2F20 X5D05 X5D10

Fig. 6. Variation of RH with degree of mixing of efuent air (0  C, 100% RH) with ambient air (red curve: 30  C, 80% RH; blue curve: 30  C, 40% RH). (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

The main difference between modeling a fog cloud and other dense gas releases is the phase change that occurs when fog is formed or dissipated. Fog formation reduces the amount of excess moisture by forming water droplets; it is also an exothermic process. Conversely, fog dissipation converts water droplets back into water vapor, while absorbing heat. A built-in model for moisture condensation/evaporation in the bulk ow, that can track the liquid phase (fog) mass, is not available in most current commercial CFD codes. Therefore, the method presented in this paper applies user dened functions to implement the phase change between water vapor and fog droplets, as dened by the psychrometric model equations. When the conditions for fog formation (i.e., RH > 100%) are predicted by the gas dispersion model, condensation of water vapor occurs and mass is transferred from water vapor to fog. However, fog is modeled as a gas-phase component (not as a liquid phase) with the same thermophysical properties as liquid water. When conditions for fog dissipation are predicted by the dispersion model (i.e., RH < 100% and F > 0), the reverse process occurs: fog droplets evaporate and mass is transferred from fog to water vapor. In both scenarios, latent

heat of vaporization is also released or absorbed according to the direction of mass transfer. This approach is computationally more efcient than a true multiphase (gaseliquid) simulation. In fact, this model allows the convective- and turbulent diffusion-driven motion of the fog to be accurately simulated, while neglecting the gravitational drift of the small fog droplets. The practical reason for concern with AAV-induced fog formation is the impact of fog on visibility, both in proximity of the vaporizers (e.g., for personnel operations) and farther away (e.g., for impact on road or maritime trafc). In fact, fog consists of a large number of minute water droplets suspended in air (ranging in size from approximately 0.3 to over 10 mm). As light passes through fog, it is scattered by the water droplets so that both the intensity of the light and the contrast of objects are reduced. An object is no longer visible when its contrast with the background can no longer be detected. Correlating fog density to visibility reduction is quite complicated, given the large number of droplets, the stochastic droplet size distribution and its dynamic variation due to condensation, evaporation and coalescence of droplets. Under the assumptions of this model (where droplet size distribution is assumed constant and equal to that of typical fog), the effect of fog concentration on visibility can be expressed according to the correlation by Arnulf, Bricard, Cur, and Veret (1957):

V 0:024f 0:65
where f is the fog density (g/m3) and V is the visibility (km). 5. Model validation

(7)

There is currently no available experimental data on fog dispersion that would allow a direct validation of the model presented in this paper. However, condence in the accuracy of the

Fig. 7. Detail of the computational mesh for the 6-AAV array fog dispersion scenario (three AAVs are visible).

778

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

Fig. 8. Temperature on a streamwise vertical plane through one row of AAVs, for 2 m/s wind speed: 1.5 m discharge elevation (top); 6 m discharge elevation (bottom).

Fig. 9. Fog isocontour with concentration 106 kg/m3 for dispersion from 1.5 m elevation discharge: 2 m/s wind speed (top); 5 m/s wind speed (bottom).

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

779

model can be obtained by separately testing the two main components of a fog dispersion model: the dispersion of a dense gas release and the psychrometric model. The dense gas dispersion capabilities of Star-CCM were tested by simulating the LNG vapor cloud dispersion experiment known as Burro 8 (Koopman et al., 1980). The Burro 8 experiment consisted of a spill of LNG onto a water pond and dispersion of the resulting vapor cloud over mostly at terrain, under low wind and a stable atmosphere. The CFD simulation of the Burro 8 scenario predicted a maximum distance to the lower ammable limit (LFL) within approximately 11% of the experimental value. Additionally, the peak gas concentration at various downwind locations predicted by the CFD model was well within a factor of 2 of the corresponding experimental data. The psychrometric model was tested by comparison with the adiabatic saturator theory (Moran & Shapiro, 2000). A CFD model of a long tube (length/diameter ratio of approximately 100) with adiabatic walls was created and two moist air streams (a cold and a warm stream) were inserted concentrically at the inlet. The mass ow-averaged temperature, relative humidity and fog concentration predicted by the CFD model at the outlet of the tube were compared with the theoretical values obtained from the psychrometric chart. A total of seven combinations of cold/warm stream values were tested. The CFD predictions were within less than 0.5% of the corresponding theoretical values for the non-saturated mixtures and within approximately 12% of the corresponding theoretical values for fog conditions.

Based on the results of the dense gas dispersion and of the psychrometric model, the fog dispersion method presented in this paper can be expected to provide acceptable estimates of the fog dispersion distances. 6. Examples of application of the fog dispersion model The dispersion of the fog cloud generated by an array of six forced-draft AAVs was simulated in order to demonstrate the application of the fog dispersion model. The AAVs were grouped in two rows of three units each, in the direction of the wind. The vaporizers measured approximately 3 m by 3 m by 12 m tall and were spaced approximately 1.5 m apart. A total of ve scenarios were examined, in which the wind speed was varied from 2 to 5 m/s (at 10 m elevation) and the elevation of the efuent discharge was varied between approximately 1.5 and 6.0 m, to evaluate the effect of both parameters on fog dispersion. The vaporizers were assumed to be shrouded, as is typical for forced-draft units. Each vaporizer was assumed to generate an air ow rate of approximately 45 kg/s, with the efuent being at saturation at 0  C. Ambient air was assumed to be at 30  C temperature and 80% relative humidity. The ve scenarios were set up as follows:  Wind speed of 5 m/s at 10 m elevation and neutral atmospheric stability (PasquilleGifford D class). Two scenarios were run, with AAV discharge elevations of 1.5 and 3 m, respectively;

Fig. 10. Streamlines for AAV efuent from 1.5 m elevation discharge: 2 m/s wind speed (top); 5 m/s wind speed (bottom). Streamlines are colour-coded according to temperature.

780

F. Gavelli / Journal of Loss Prevention in the Process Industries 23 (2010) 773e780

 Wind speed of 2 m/s at 10 m elevation and stable atmosphere (PasquilleGifford F class). Three scenarios were run, with AAV discharge elevations of 1.5, 3, and 6 m, respectively. The computational domain measured approximately 300 m wide by 400 m long by 30 m high and was discretized using a hexahedral mesh with prismatic boundary layer cells at the walls. As shown in Fig. 7, a ne mesh with minimum cell size of approximately 0.15 m was applied to a volume between the ground and approximately 0.6 m above the AAV discharge, enveloping the region where fog is expected. A variable size mesh was used to ll the rest of the domain, with cell size growing progressively up to approximately 9.2 m. A total of approximately 530,000 cells were used in the simulation. The realizable keepsilon model (Shih, Liou, Shabbir, Yang, & Zhu, 1994) was used for turbulence closure and the simulations were run using the steady solver. The inlet boundary condition proles for velocity, temperature, turbulent kinetic energy and dissipation rate were obtained from the MonineObukhov theory (Golder, 1972). The ground was assumed to be at a constant temperature of 30  C. The maximum downwind distance reached by the fog cloud was used as the quantitative term of comparison between the ve scenarios. As shown in Table 1, higher wind speed and higher discharge elevation tend to accelerate the dissipation of the fog cloud. The effect of discharge elevation is particularly strong once the discharge elevation is raised above 3 m (scenario X2F20). The effect of discharge elevation is also evident in Fig. 8, which shows the air temperature on a streamwise vertical plane through a row of AAVs for low wind speed scenarios. In the top image (1.5 m discharge elevation), the cold temperature efuent lls the gap between the AAVs and the ground, creating a cold core shielded from the warmer ambient air. In the bottom image (6 m discharge elevation), the warm ambient air penetrates the region below the AAV discharge, increasing the mixing surface between the two uids and resulting in faster dissipation of the fog cloud. Fig. 9 shows the fog cloud isocontours (at a fog concentration of 106 kg/m3) for the low and high wind speed scenarios (top and bottom image, respectively) and an AAV discharge elevation of 1.5 m. The behavior of the fog cloud in the two cases is noticeably different. In the low wind case, the fog cloud extends approximately 50 m laterally (crosswind) on both sides of the AAV array, as well as approximately 10 m upwind; the cloud remains close to the ground, never stretching more than approximately 3 m above the ground. The streamlines plot in Fig. 10 (top) shows how the AAV efuent impinges on the ground and initially spreads radially; the wind affects the efuent jet dispersion only farther away from the AAV towers. In the high wind speed case, instead, the wind is affecting the AAV efuent even in the region directly beneath the towers: the jet streamlines (Fig. 10, bottom) are immediately deected and impinge on the ground farther downwind. As shown in Fig. 9 (bottom), there is minimal crosswind spread and no upwind dispersion of the fog cloud. The stronger wake formed downwind of the AAV array lifts up the fog cloud, which reaches

a maximum thickness of approximately 9 m immediately downwind of the towers, before dissipating a short distance away. 7. Conclusions Ambient air vaporizers (AAVs) are widely used to regasify liqueed industrial gases, which are liqueed for transport and storage. Depending on the conditions (temperature and relative humidity) of ambient air and AAV efuent, the potential exists for the formation of fog as the two uids mix with each other. This possibility has raised some regulatory and environmental concerns that the fog cloud may impact human activities in the vicinity of the AAV arrays. A method to quantitatively predict the formation, advection and dissipation of the AAV-induced fog using a commercial CFD model was developed and applied to an array of six forced-draft AAVs. The discharge from an array of six forced-draft AAVs was used as an example to demonstrate the application of the fog dispersion simulation method. The example evaluated the effects on fog dispersion of AAV discharge elevation and of wind speed. Consistent with the experience gathered from other dense gas releases, higher wind speed was shown to accelerate the dissipation of the fog cloud. An increase in the elevation of the AAV discharge resulted in faster mixing of the efuent with ambient air and, as a result, shorter fog dispersion distances. References
Arnulf, A., Bricard, J., Cur, E., & Veret, C. (1957). Transmission by haze and fog in the spectral region 0.35 to 10 microns. Journal of the Optical Society of America, 47, 491e498. Barenbrug, A. W. (1974). Psychrometry and psychrometric charts (3rd ed.). Britter, R. E. (1989). Atmospheric dispersion of dense gases. Annual Review of Fluid Mechanics, 21, 317e344. Carey, G. F. (1997). Computational grids: Generation, adaptation, and solution strategies (pp. 150e153). Taylor and Francis. Gavelli, F., Bullister, E., & Kytomaa, H. (2008). Application of CFD (uent) to LNG spills into geometrically complex environments. Journal of Hazardous Materials, 159, 158e168. Golder, D. (1972). Relations among stability parameters in the surface layer. Boundary-Layer Meteorology, 3, 47e58. Koopman, R., Baker, J., Cederwall, R. T., Goldwire, H. C., Hogan, W. J., Kamppinen, L. M. et al. (1980). Burro series data report. LLNL/NWC 1980 LNG spill tests, Lawrence Livermore National Laboratories. Report no. UCID-19075. Luketa-Hanlin, A., Koopman, R., & Ermak, D. (2007). On the application of computational uid dynamics codes for liqueed natural gas dispersion. Journal of Hazardous Materials, 140, 504e517. McQuiston, F. C., Parker, J. D., & Spitler, J. D. (2000). Heating ventilation and air conditioning. Analysis and design (5th ed.). Wiley. Moran, M. M., & Shapiro, H. N. (2000). Fundamentals of engineering thermodynamics (3rd ed.). Wiley. Shah, K., Wong, J., & Minton, B. (2008). Considerations for ambient air based technologies for LNG regasication terminals. In Proc. 2008 AIChE Spring meeting. Shih, T. H., Liou, W. W., Shabbir, A., Yang, Z., & Zhu, J. (1994). A new ke3 Eddy viscosity model for high Reynolds number turbulent ows e model development and validation. NASA TM 106721. Turner, D. B. (1964). A diffusion model for an urban area. Journal of Applied Meteorology, 3, 83e91.

You might also like