lable at ScienceDirect

Atmospheric Environment 109 (2015) 262e271

Contents lists avai

Atmospheric Environment

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

The near-source impacts of diesel backup generators in urbanenvironments

Zheming Tong, K. Max Zhang*

Sibley School of Mechanical and Aerospace Engineering, Cornell University, Ithaca, NY 14853, USA

h i g h l i g h t s

� We investigated the local air quality impacts of diesel backup generators (BUGs).� The impacts strongly depend on urban configurations and meteorological conditions.� The presence of a tall upwind or downwind building forms severe PM hotspots.� Demand response programs with diesel BUGs can result in local air quality problems.� The siting requirements of BUGs need to be revisited to consider those impacts.

a r t i c l e i n f o

Article history:Received 20 October 2014Received in revised form26 February 2015Accepted 9 March 2015Available online 11 March 2015

Keywords:Distributed generationEmergency generatorPlume dispersionMicrometeorologyAtmospheric stabilityCFD

* Corresponding author.E-mail address: kz33@cornell.edu (K.M. Zhang).

http://dx.doi.org/10.1016/j.atmosenv.2015.03.0201352-2310/© 2015 Elsevier Ltd. All rights reserved.

a b s t r a c t

Distributed power generation, located close to consumers, plays an important role in the current andfuture power systems. However, its near-source impacts in complex urban environments are not wellunderstood. In this paper, we focused on diesel backup generators that participate in demand response(DR) programs. We first improved the micro-environmental air quality simulations by employing ameteorology processor, AERMET, to generate site-specific boundary layer parameters for the Large EddySimulation (LES) modeling. The modeling structure was then incorporated into the CTAG model toevaluate the environmental impacts of diesel backup generators in near-source microenvironments. Wefound that the presence of either tall upwind or downwind building can deteriorate the air quality in thenear-stack street canyons, largely due to the recirculation zones generated by the tall buildings, reducingthe near-stack dispersion. Decreasing exhaust momentum ratio (stack exit velocity/ambient wind ve-locity) draws more exhaust into the recirculation zone, and reduces the effective stack height, whichresults in elevated near-ground concentrations inside downwind street canyons. The near-ground PM2.5

concentration for the worst scenarios could well exceed 100 mg m�3, posing potential health risk topeople living and working nearby. In general, older diesel backup generators (i.e., Tier 1, 2 or older)without the up-to-date emission control may significantly increase the pollutant concentration in thenear-source street canyons if participating in DR programs. Even generators that comply with Tier-4standards could lead to PM hotspots if their stacks are next to tall buildings. Our study implies thatthe siting of diesel backup generators stacks should consider not only the interactions of fresh air intakeand exhaust outlet for the building housing the backup generators, but also the dispersion of exhaustplumes in the surrounding environment.

© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Small distributed power generation is becoming more populardue its flexibility and efficiency compared with central power

generation (Pepermans et al., 2005). These units are typicallylocated in populated urban areas with relatively short stack heights.Since they are closer to consumers, their environmental impactshave become a concern despite the benefits. Several studies haveevaluated the air quality impact from distributed generation(Greene and Hammerschlag, 2000; Jing and Venkatram, 2011;Heath et al., 2006; Strachan and Farrell, 2006; Carreras-Sospedraet al., 2010). However, very few studies have examined the effects

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271 263

of complex urban environments such as street canyons on the near-source air quality impacts. This paper aims to bridge this gap.

Diesel backup generators, often referred as “standby generators”or “emergency generators”, are one type of distributed generation.Their primary purpose is to preserve essential facility functions inthe event of a loss of grid power or for situations that threaten thefacility, such as fire pump use during a fire (NESCAUM, 2012). Thosegenerators can also operate during periods of peak electricity de-mand, increasing grid reliability and supporting the electricitydelivery systems (Gilmore et al., 2006). For example, it is estimatedthat over 1,000 MW of backup generation capacity has beeninstalled in New York City (NESCAUM, 2003), and the New YorkIndependent System Operators (NYISO) allows backup generatorsto participate in its demand response (DR) programs (Gilmore et al.,2010). Although there are several benefits of using diesel backupgenerators for DR, many of them have non-trivial air emissions(Gilmore et al., 2006; Shah et al., 2006). The U.S. EnvironmentalProtection Agency (USEPA) first introduced Tier 1 emission regu-lation for non-road diesel generators in 1996, and the permittedlevels of nitrogen oxides (NOx) and particulatematter (PM), the twomain pollutants from diesel engines, have gone down significantlysince then (USEPA, 2014). However, a large percentage of dieselbackup generators that are in use are Tier 1, Tier 2 or older, whichhave considerably higher emission rates than those of the latestmodels (Zhang and Zhang, 2015). Due to the short operating time,the annual emissions of NOx and PM from those generators aresmall. However, DR programs usually take place during hazy, hotand humid summer days. As a consequence, the emissions of dieselbackup generators participating in DR programs may contribute toregional ozone pollution and local PM hotspots (Zhang and Zhang,2015).

To evaluate the local air quality impact of distributed generationin urban neighborhoods, it is necessary to accurately simulateplume dispersion in microenvironments where exhaust mo-mentum/buoyancy, surrounding structures and micrometeorologyplay significant roles. Gaussian plume models are commonly usedto evaluate local air quality. However, these models cannotexplicitly resolve the complex flow field generated by near-sourcestructures such as street canyons (Gilmore et al., 2006; Blockenet al., 2008), which can significantly modify plume dispersion(Robins and Castro, 1977). Computational fluid dynamics (CFD)models, on the other hand, are more capable of capturing thecomplex unsteady fluid dynamics and dispersion, but at a highercomputing cost. Large Eddy Simulation (LES) is employed formodeling turbulent flow and dispersion in this study. It is anappropriate model for the purpose of our study as it resolves thelarge-scale unsteady motions and requires modeling only thesmall-scale, unresolvable turbulent motion, which is less influ-enced by the physical boundary conditions (Xie and Castro, 2006).Several LES studies were conducted on real urban geometry, andgeneral modeling guidelines were developed (Tseng et al., 2006;Tamura, 2008; Gousseau et al., 2011). It is found that thermalstratification resulted from atmospheric stability critically affectsnear-surface concentrations of pollutants (Sini et al., 1996; Woodand Jarvl, 2012). Therefore, site-specific boundary layer parame-ters such as friction velocity, stability length, and sensible heat fluxneed to be assessed for micro-environmental plume dispersionstudies.

The first objective of this study is to introduce an improvedapproach to simulate spatial variations of pollutants in a near-source urban microenvironment by employing a meteorologyprocessor AERMET to generate site-specific boundary layer pa-rameters as boundary conditions (USEPA, 2004a). This enhancesthemodeling capability in resolvingmicro-scale air quality at urbanneighborhood-scale (Wang et al., 2013; Tong et al., 2011; Steffens

et al., 2012). The second objective is to evaluate the environ-mental impact of diesel backup generators in near-sourcemicroenvironments.

The first part of the paper focuses on evaluating the interactionsof plume dispersion and building downwash against existing windtunnel experiments. The second part presents case studies toquantitatively evaluate the local impacts of NOx and PM2.5 emis-sions from diesel backup generators. A series of LES simulations areconducted under various atmospheric stability conditions and ur-ban configurations.

2. Model description

The Comprehensive Turbulent Aerosol Dynamics and GasChemistry (CTAG) model is designed to resolve the flow fieldincluding turbulent reacting flows, aerosol dynamics, and gaschemistry in complex urban environments. In this paper, weexpand the existing CTAG model by incorporating the micromete-orology component in order to simulate thermal stratification andbuoyancy effects that strongly influence the plume dispersion inmicroenvironments (Section 4.1). A full description of the model'stheoretical background and implementation is presented in ourprevious work and Supporting Information (Tong et al., 2012;Steffens et al., 2012; Wang and Zhang, 2009, 2012; Wang et al.,2011, 2013; Steffens et al., 2013, 2014). A short description is pre-sented here. LES is employed to resolve the unsteady turbulent flowfield. In LES, a low-pass filtering operation is performed so that theresulting velocity field ~ui can be resolved on a relatively coarse grid.A dynamic subgrid model is chosen, which allows the Smagorinskyconstant to vary in space and time (Germano et al., 1991). Theconstant is dynamically computed based on the information pro-vided by the resolved scales of motion. Logarithmic wall function isapplied to the near-wall region as it is computationally impracticalto resolve every viscous sublayer in a large domain (Launder andSpalding, 1974). Turbulence at urban environment is dominatedby large-scale motion generated by scales comparable to the size ofbuildings and street canyons (Xie and Castro, 2006). Capturing theprecise surface drag of each building surface is exceedinglyexpensive.

3. Model evaluations

We evaluate the LES model against two wind tunnel datasets onturbulenct flow and plume around surface-mounted cubes,respectively. The former is from ERCOFTAC (European ResearchCommunity on Flow, Turbulence and Combustion), and the latter isfrom USEPA (Martinuzzi and Tropea, 1993; Thompson, 1993). Theevaluations serve as a solid foundation for simulating plumedispersion in realistic urban environments.

3.1. ERCOFTAC

3.1.1. ExperimentFor the flow field experiment from ERCOFTAC, the Reynolds

number based on the cube height is ~40,000. The turbulent flowfield around surface-mounted prismatic obstacles was character-ized using the crystal violet, oil-film and laser-sheet visualizationtechniques. The experiment was performed in an open, blower-type air channel. The dimensions of the channel are390 cm� 60 cm x 5 cm. The boundary layer was created at the inletin order to obtain fully developed flow condition.

3.1.2. SimulationsThe schematic drawing for the flow experiment is shown in

Fig. 1a. The inlet boundary condition is obtained from the

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271264

experimental measurement including mean flow and fluctuationcomponents. The time-dependent feature of the inlet turbulenceprofile is simulated by the vortex method, where random vorticesin the inlet flow plane for the wall-normal components aregenerated giving a spatial correlation (Mathey et al., 2006). Sym-metry boundary conditions are applied for the two sides as slipwalls with zero shear. Outflow boundary condition is specified atthe end of the domain. The computational domain contains about1.7 million cells. The grids are more refined near the solid cube andstack. Prism layer mesh is applied near the wall in order to reducenumerical diffusion.

3.1.3. ResultsFig. 2 depicts a comparison of predicted and measured time-

averaged streamwise flow velocity profiles. In general, the perfor-mance of LES model is adequate. Slight discrepancies are observedat the cube wall (z/H ¼ 1) and floor (z/H¼ 0) due to semi-empiricalwall function employed. H is defined as the height of the cube. Fig. 3demonstrates a satisfactory agreement between the predicted andmeasured time-averaged velocity component in the streamwisedirection on the symmetry plane at two different heights (z/H¼ 0.5and z/H ¼ 0.9). The LES model generally captures the trend that themean flow velocity decreases to zero at the building wall due to no-slip condition and gradually recovers to upstream velocity at aboutx/H ¼ 8. However, the model overpredicts the negative mean ve-locity near the leeward face between x/H ¼ 4 and 5.

3.2. Plume dispersion

3.2.1. ExperimentThe plume dispersion data are obtained from a separate

experiment fromUSEPA, which investigated how the aerodynamicsaround buildings with different stack heights (SH) influence thedispersion of pollutants. Two scaled building widths (15 cm and60 cm) are chosen for evaluation as shown in Fig. 4. The experimentwas conducted in the USEPAmeteorological wind tunnel with a testsection of 3.7 m (W) � 2.1 m (H) � 18 m (L). A neutral boundarylayer was generated by tripping the flow with a high fence at theentrance. The Reynolds number based on the building height andthe approaching flow velocity at the same height is ~32,400, whichis well above the critical Reynolds number (Thompson, 1993).Tracers were released at different SHs both above and below thebuilding height, and the ground-level concentrations along thetunnel centerline parallel to the flow direction were measured.However, the flow fields were not characterized. Concentrationsand distances in the experiment have been normalized in order to

Fig. 1. a) Schematic of the numerical setup for flow field evaluation b) Schematic of nSH2 ¼ 150 mm, SH3 ¼ 225 mm, SH4 ¼ 300 mm. H is defined as the height of the cube.

make appropriate comparisons between the scale model created inthe wind tunnel and full scale, real-world scenarios. Distances havebeen normalized by the building height. Concentrations have beennormalized according to the following formula,

c ¼ CUHH2

Q;

where H is the building height, UH is the approach flow speed at thebuilding height, and Q is the source emission rate.

3.2.2. SimulationsThe schematic drawing for the plume dispersion experiment is

shown in Fig. 1b. The locations of inlet, outlet and symmetryboundary conditions are similar to the ERCOFTAC simulationsexcept the 4 SHs (75, 150, 225, 300 mm above the floor) at the inletwall where the plumewas released. Zero diffusive flux is applied toevery other solid wall. The size of the computational domain foreach case contains about 2.2 million cells. The grids are morerefined near the solid cube and stack. Prism layer mesh is appliednear the wall in order to reduce numerical diffusion.

3.2.3. ResultsFig. 5 compares the predicted and measured tracer concentra-

tions for the 15 cm and 60 cm buildings along the centerline of thedomain. x/H ¼ 0 is at the windward face of the cube. Overall, theLES model adequately captures the general trends and peak con-centrations. The performance of LES is satisfactory at 4 SHs, thoughthe model overpredicts the centerline concentration prior to thepeak location for the 15 cm building for the SH ¼ 225 mm case.

4. Diesel backup generator case studies

4.1. Numerical setup

To evaluate local air quality impacts due to emissions fromdiesel backup generators, the size of the modeling domain needs besufficiently large to include nearby buildings and other structures.The height of the domain should be tall enough to account forplume rise, and avoid any blocking effect and unphysical flow ac-celeration (Tominaga et al., 2008). Fig. 6 shows the modeling do-mains. The baseline building (i.e., where the backup generator ishoused) and stack parameters are based on a realistic urban area inNew York City. The size of the domain is about1,200 � 1,000 � 150 m. It is meshed with 4.5 million unstructuredelements with prism layers near the wall. Grid independency study

umerical setup for the plume dispersion evaluation. SH1 (Stack Height) ¼ 75 mm,

Fig. 2. Time-averaged streamwise velocity profiles at various x/H as indicated.

Fig. 3. Streamwise velocity profile on the symmetry plane at two heights z/H ¼ 0.5, 0.9.

Fig. 4. Schematics of the building used in the plume dispersion experiment. Left: 15 cm building; Right: 60 cm building.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271 265

is conducted to ensure the results are independent of the meshresolution.

Vertical wind and temperature profiles (>100 m) are usually

unavailable for most cities. On the other hand, the quality ofmicroscale simulation relies on those profiles specified at theboundaries (Tominaga et al., 2008). Therefore, a realistic estimation

Fig. 5. a) Normalized ground-level centerline concentration (c) for the 15 cm building at SH ¼ 75, 150, 225, 300 mm; b) Normalized ground-level centerline concentration (c) forthe 60 cm building at SH ¼ 75, 150, 225, 300 mm.

Fig. 6. a) Geometry of the diesel backup generator site, b) Cut plane that intersects the stack and parallel to xez plane. Vertical lines that are 30, 100, 300, 500 m downwind from thestack are shown.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271266

Table 1Stack, meteorology, and building parameters for a 1000 kW diesel backup generatorat steady-state operation.

Parameters Values References

Exhaust velocity 15 m s�1 Petersen et al. (2002)Exhaust temperature 650 Ka CaterpillarStack height 3.1 m from rooftop Petersen et al. (2002)Stack inside diameter 0.77 m Petersen et al. (2002)NOx emission factor 10.6 g kWh�1 Zhang and Zhang (2015)PM2.5 emission factor 0.5 g kWh�1 Greene and Hammerschlag

(2000); Gilmore et al. (2006);USEPA (2006)

Emission standards Mix of Tier 1, 2and Pre-Tier

Zhang and Zhang (2015)

Stability Unstable, neutral,and stable

N/A

Urban configuration 4 cases shownin Fig. 8

N/A

a Temperature drop over exhaust duct considered.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271 267

of inflow profiles becomes critical for LES simulations at this scale.Domain of this size locates in the surface layer of planetaryboundary layer, which is amenable to Monin-Obukhov similaritytheory employed in this study (Monin and Obukhov, 1954). Thus, asemi-empirical approach is introduced here based on local weatherdata and Monin-Obukhov similarity theory. A meteorological pro-cessor, AERMET, developed by USEPA is employed in our study(USEPA, 2004a; Cimorelli et al., 2005). The friction velocity (u*) andMonin-Obukhov length (L) are computed from the quadratic so-lution for stable condition, and iterative method for unstable con-dition (Hanna and Chang, 1993; Perry, 1992). Sensible heat flux (H)is estimated based on energy balance approach developed by Oke(1988). Hourly surface data are obtained from the weather stationat LaGuardia Airport, and upper air data from Brookhaven, NY.Surface characteristics at a resolution of 1 km including surfaceroughness length z0, albedo, and Bowen ratio at the specifiedlocation are computed using AERSURFACE developed by USEPA,which extracts data from U. S. Geological Survey (USGS) NationalLand Cover Data archives (USEPA, 2008). An in-house script isdeveloped to generate inflow profiles for LES simulations based onMonin-Obukhov similarity theory, and boundary layer parametersfrom AERMET (u*, L, H, and z0). Detailed discussions can be found inthe Supporting Information.

The fully developed wind velocity and temperature profiles areapplied at the west boundary as inlet, which is the dominant winddirection and there are multiple street canyons downwind of thestack. Symmetry boundary condition is applied on the north andsouth sides of the domain as slip walls with zero shear. At the flowoutlet (east boundary), zero diffusion flux of all flow variables isspecified. Directly simulating heat transfer from ground andbuildings with LES requires resolving the thermal layer at everybuilding surface, which is very computationally expensive, andthere are no thermal-wall models available yet (Xie et al., 2013).Thus, an alternative approach is employed. AERMET is used tocompute thermally stratified approaching flow for the entire site.The computed profiles are then applied at the inlet boundary of themodeling domain. At the building and ground surfaces, adiabaticand no-slip conditions are applied.

Exhaust particles in diesel backup generator emissions mostlyconstitute of those less than 1 mm (USEPA, 2006). The volumefraction of the particulate phase is low, and submicron particleshave small relaxation time, which are able to closely follow thefluid streamline. Therefore, the flow falls into the dilute gasesolidflow regime. In addition, we treat primary PM2.5 as a tracer species,i.e., assuming that gas/particle partitioning near stacks will notsignificant change the primary PM2.5 mass near the sources. Thisassumption is subject to future investigation. Thereby, both NOXand PM2.5 concentration fields are solved by convection-diffusionequation. Surface deposition of PM2.5 at building surfaces isaccounted for by imposing a mass flux based on mass-weighteddeposition velocity from the literature (Petroff and Zhang, 2010),discussed in the Supporting Information. In the modeling domain,the diesel backup generator is treated as the only source, and nobackground concentration is considered.

4.2. Stack parameters

The diesel backup generators are typically installed in thebasements, and their exhaust stacks are located on the rooftops.The exhaust stack is designed to be sufficiently far from the fresh airintake of the same building housing the diesel backup generator.The rooftop stacks of diesel backup generators are typically muchshorter than the ones installed for power plants due to aestheticsand local building code (Lomas, 2007; Carter et al., 2005; Porter,2006). We create a baseline case, where a 3.1 m rooftop stack is

located on a two-story building upwind of several street canyons.The stack height is above the roof recirculation zone estimated byASHRAE in order to avoid plume entrainment on the rooftop(ASHRAE, 2003). The parameters for a typical rooftop diesel backupgenerator stack are shown in Table 1. Diesel backup generators areable to reach full load within 10 s (Wroblewska, 2011). Thus onlysteady-state stack parameters are considered. Both NOx and pri-mary PM2.5 emission factor are estimated from the mix of Tier 1,Tier 2, and Pre-Tier diesel generators. Two generator sizes (500 kWand 1000 kW) are tested in this study.

4.3. Results and discussions

The vertical profiles of NOx concentration (ppmv) are shown inFig. 7a. The corresponding weather data are chosen from hoursunder different stability classes in July, 2013. The temperature andwind profiles for these simulations are described in the SupportingInformation. The concentration profiles give a general idea of theplume trajectory as a function of distance and stability. The atmo-spheric stability plays an important role in plume dispersion. Theunstable condition results in the highest plume trajectory andeffective stack height because of the vertical movement of airparcels and the additional buoyance created by thermal stratifica-tion. In comparison, the stable condition leads to significantly lowerplume trajectory and buoyance because it resists vertical mixing.The trajectory of neutral condition stays in between stable andunstable conditions.

At 30 m from the stack, peak concentrations of NOx for threestability classes are in the range from 7 to 10 ppmv, and decreasebelow 0.5 ppmv at 500 m downwind. The peak primary PM2.5concentrations of the three stability classes reaches the range be-tween 600 and 1,100 mg m�3 at 30 m downwind (z~18 m) of thestack, and dilute to 15e40 mg m�3 range at 500 m. More verticaldispersion and near-ground impacts are observed at x ¼ 300,500 m as the plume travels downwind shown in Fig. 7b.

Surrounding street canyon configurations play an importantrole on near-stack plume dispersion. For further investigations,three additional configurations are created to compare against thebaseline (Case A). Case B represents the condition where theadjacent downwind building is significantly taller (3.5HSH) than thebackup generator building, where HSH is the sum of height of thebaseline backup generator building and stack height. Case C is thecondition with a tall building upwind of the stack (2.5HSH), whichcreates a recirculation zone between the upwind and downwindbuilding of the stack. Case D presents a scenario where the upwindand downwind buildings are the same as Case C, but the new

Fig. 7. a) Vertical profiles of NOx concentrations of backup diesel generator in ppmv b) Vertical profiles of PM2.5 concentrations of backup diesel generator in mg m�3 at 30 m, 100 m,300 m, and 500 m away from the stack.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271268

backup generator building is taller than the upwind building andoutside the recirculation zone (3xHSH). The rest of the stack pa-rameters are the same for all four cases.

We have evaluated the meteorological conditions for the hours(typically in the afternoons) when NYISO emergency-based DRprograms were called from 2011 to 2013 (NYISO, 2013). Unstableatmospheric conditions are found to persist for every DR eventhour during this period. The specific boundary meteorologicalconditions used in our simulations are determined based on aparticular hour representing the median temperature (i.e., 34 �C) ofall DR event hours on July 18, 2013. The wind velocity at thereference height was about 4m s�1. Fig. 8 displays the contour plotsof primary PM2.5 concentration for each configuration within200m from the stack. A contour plot of the entire domain for Case Cis included in the Supporting Information.

At the baseline configuration (Case A), no near-ground con-centration in the adjacent street canyons is observed due to suffi-cient HSH and the unstable atmosphere that enhances verticaldispersion. The plume rises higher above the ground as it travelsdownwind. In Case B, the plume trajectory is pushed higher thanthe baseline case due to the upward flow deflection near thewindward wall. The tall downwind building creates a large recir-culation zone behind it. The resulting wake zone behind thebuilding slows down the mean flow and creates a well-mixed re-gion, which elevates the near-ground concentration. For Case C, thestack is located inside the recirculation zone generated by the tallupwind building. In this wake zone, the stack plume is drawn

downward and sideways, which considerably reduces near-stackdilution. This building downwash effect of the plume significantlyraises the primary PM2.5 concentration above 200 mgm�3 inside thecanyon between the upwind and downwind buildings. In Case D,the building downwash of the plume between the upwind anddownwind buildings is substantially reduced by locating the stackoutside the recirculation zone. However, the primary PM2.5 con-centration in the street canyons further downwind of the stack isstill around 20 mgm�3 due to the largewake zone created by the talldiesel backup generator building.

In order to understand the effects of ambient wind velocity onplume dispersion, we conduct a sensitivity study on the upcomingwind speed using the configuration in Case C, as it creates the mostdownwash effect among the four configurations illustrated in Fig. 8.The exhaust momentum ratio M (¼ stack exit velocity/ambientwind velocity) is employed here for the three test cases shown inFig. 9 (Petersen et al., 2002; Blocken et al., 2008). M¼ 3.8 from CaseC is chosen as a new baseline for the comparison. Comparing thethree ratios, it is evident that decreasing M reduces the downwindvertical mixing of the plume, and causes the plume to travel closerto the ground level. As a result, it leads to lower effective stackheight and elevated pollutant concentrations in the downwindstreet canyons. When M is increased to 7.6, the primary PM2.5concentration inside the street canyon ranges from 0 to 30 mg m�3.When M decreases to 1.9, the primary PM2.5 concentration insidethe street canyon increases significantly to over 200 mg m�3 as theplume is pulled more downward into the recirculation zone. Since

Fig. 8. Schematics of 4 configurations and corresponding contour plots of primary PM2.5 concentration in mg m�3. Case A is the baseline configuration.

Fig. 9. Contour plots of primary PM2.5 concentration in mg m-3 for three exhaust momentum ratio M (¼stack exit velocity/ambient wind velocity) indicated. Stack exit velocity iskept constant for the three cases. M ¼ 3.8 from Case C is employed as a new baseline for comparison. Ambient wind velocity for the M ¼ 7.6 case is 2 m s�1, and 8 m s�1 for theM ¼ 1.9 case.

Fig. 10. Primary PM2.5 concentration contour plots (mg m�3) of Case C for a) 1000 kW diesel backup generator b) 500 kW diesel backup generator.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271 269

the stack is located inside the wake zone created by the upwindbuilding, increasing ambient wind velocity does not enhance thenear-stack dilution.

The environmental impact of a smaller 500 kW diesel backupgenerator is also evaluated for Case C. The same emission factor forPM2.5 (in g kWh�1) as presented in Table 1 is used. All other stackparameters are the same except slightly greater exhaust

temperature (Caterpillar). The contour plot of the primary PM2.5concentration for the 500 kW backup generator is shown in Fig. 10.The plume dispersion patterns are similar to 1,000 kW case. Lowernear-ground concentration in the downwind street canyons isobserved.

The latest USEPA Tier 4 emission standards for stationary dieselengine were implemented in 2014 (USEPA, 2004b). The emission

Fig. 11. Contour plots of a) NOx concentration in ppmv and b) PM2.5 concentration mg m�3 based on Tier 4 standards based a 1,000 kW diesel backup generator.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271270

rates are significantly reduced than the ones in Table 1. NOx emis-sion factor is 2.16 g kWh�1, and PM2.5 emission factor is 0.04 gkWh�1, which are almost 5 and 12 times less than the ones fromTier 1 and older, respectively (Zhang and Zhang, 2015). Based onthese emission factors, Case C is re-simulated shown in Fig. 11.Reduced NOx concentration in the street canyon between upwindand downwind building is around 30 ppbv. Primary PM2.5 con-centration is decreased below 30 mg m�3 inside the street canyon.Although the reduction is significant, the near-source impact is stillnot negligible during operation hours.

5. Conclusion

Backup diesel generators are often allowed to participate indemand response (DR) programs during peak demand periods. Inthis study, we investigate the spatial variations of air pollutantsnear a diesel backup generator stack by employing an improvedapproach for micro-environmental simulations, in which a mete-orology processor AERMET is adopted to generate site-specificboundary layer parameters. According to our analysis, the near-source air quality impacts of diesel backup generator emissionsdepend strongly on urban configurations and meteorological con-ditions. We analyze four street canyon configurations. The worst-case scenario is identified (Case C in Fig. 9) where the stack islocated inside the recirculation zone created by a tall upwindbuilding. In this case, the plume is drawn downward and sideways,reducing the near-stack dispersion and leading to elevated con-centration inside adjacent street canyon. The near-source PM2.5concentration could well exceed 100 mg m�3 even under unstableatmospheric conditions (i.e., the conditions when DR events arecalled). In Case B (with a tall downwind building) and Case D (withthe backup generator building taller than surrounding buildings),the concentrations in the street canyon immediately adjacent to thestack are small, but elevated primary PM2.5 concentrations appearin street canyons further downwind. Decreasing the exhaust mo-mentum ratio (i.e., stack exit velocity/ambient wind velocity) drawsmore plumes into the recirculation zone, and reduces the effectivestack height, which results in an elevated near-ground concentra-tion inside downwind street canyons. Based on our study, the stacklocation should be carefully selected based on the surroundingenvironment for backup generators participating in DR programs.Otherwise, the environmental impact could potentially become anunintended consequence of DR programs, as they are traditionallyperceived as clean resources for power systems.

Acknowledgement

The authors appreciate the primary funding support from theNew York State Energy Research and Development Authority(NYSERDA), and supplemental funding support from Electric PowerResearch Institute (EPRI). The authors would also like to thank Drs.Steven Perry and David Heist at USEPA for sharing the experimentaldata described in the paper.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.atmosenv.2015.03.020.

References

ASHRAE, 2003. Application Handbook (Chapter 44).Blocken, B., Stathopoulos, T., Saathoff, P., Wang, X., 2008. Numerical evaluation of

pollutant dispersion in the built environment: comparisons between modelsand experiments. J. Wind Eng. Ind. Aerodyn. 96 (10e11), 1817e1831.

Carreras-Sospedra, M., Vutukuru, S., Brouwer, J., Dabdub, D., 2010. Central powergeneration versus distributed generation e an air quality assessment in theSouth Coast Air Basin of California. Atmos. Environ. 44 (26), 3215e3223.

Carter, J.J., Petersen, R.L., Cochran, B., 2005. Designing exhaust systems to minimizeenergy costs. ASHRAE J. 47 (7), 18e22.

Caterpillar, Technical data Diesel Generator Set CAT 3512B-1500.Cimorelli, A.J., Perry, S.G., Venkatram, A., Weil, J.C., Paine, R.J., Wilson, R.B., Lee, R.F.,

Peters, W.D., Brode, R.W., 2005. AERMOD: a dispersion model for industrialsource applications. Part I: general model formulation and boundary layercharacterization. J. Appl. Meteorol. 44 (5), 682e693.

Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale eddyviscosity model. Phys. Fluids A: Fluid Dyn. (1989e1993) 3 (7), 1760e1765.

Gilmore, E.A., Lave, L.B., Adams, P.J., 2006. The costs, air quality, and human healtheffects of meeting peak electricity demand with installed backup generators.Environ. Sci. Technol. 40 (22), 6887e6893.

Gilmore, E.A., Adams, P.J., Lave, L.B., 2010. Using backup generators for meeting peakelectricity demand: a sensitivity analysis on emission controls, location, andhealth endpoints. J. Air & Waste Manag. Assoc. 60 (5), 523e531.

Gousseau, P., Blocken, B., Stathopoulos, T., van Heijst, G.J.F., 2011. CFD simulation ofnear-field pollutant dispersion on a high-resolution grid: a case study by LESand RANS for a building group in downtown Montreal. Atmos. Environ. 45 (2),428e438.

Greene, N., Hammerschlag, R., 2000. Small and clean is beautiful: exploring theemissions of distributed generation and pollution prevention policies. Electr. J.13 (5), 50e60.

Hanna, S.R., Chang, J.C., 1993. Hybrid plume dispersion model (HPDM) improve-ments and testing at three field sites. Atmos. Environ. Part A. General Top. 27(9), 1491e1508.

Heath, G.A., Granvold, P.W., Hoats, A.S., Nazaroff, W., 2006. Intake fraction assess-ment of the air pollutant exposure implications of a shift toward distributedelectricity generation. Atmos. Environ. 40 (37), 7164e7177.

Jing, Q., Venkatram, A., 2011. The relative impacts of distributed and centralizedgeneration of electricity on local air quality in the South Coast Air Basin ofCalifornia. Energy Policy 39 (9), 4999e5007.

Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent flows.Comput. Methods Appl. Mech. Eng. 3 (2), 269e289.

Lomas, K.J., 2007. Architectural design of an advanced naturally ventilated buildingform. Energy Build. 39 (2), 166e181.

Martinuzzi, R., Tropea, C., 1993. The flow around surface-mounted, prismatic ob-stacles placed in a fully developed channel flow (Data Bank Contribution).J. Fluids Eng. 115 (1), 85e92.

Mathey, F., Cokljat, D., Bertoglio, J.P., Sergent, E., 2006. Assessment of the vortexmethod for large eddy simulation inlet conditions. Prog. Comput. Fluid Dyn. AnInt. J. 6 (1), 58e67.

Monin, A.S., Obukhov, A.M., 1954. Basic laws of turbulent mixing in the ground layerof the atmosphere. Trans. Geophys. Inst. Akad. Nauk. 24 (151), 163e187.

NESCAUM, 2003. Stationary Diesel Engines in the Northeast: an Initial Assessmentof the Regional Population, Control Technology Options and Air Quality PolicyIssues. Northeast States for Coordinated Air Use Management.

NESCAUM, 2012. Air Quality, Electricity, and Back-up Stationary Diesel Engines inthe Northeast. Northeast States for Coordinated Air Use Management.

NYISO, 2013. Annual Report to the Federal Energy Regulatory Commission on theNYISO Demand Side Management Programs. Independent System Operator,New York.

Oke, T.R., 1988. Street design and urban canopy layer climate. Energy Build. 11 (1e3),103e113.

Z. Tong, K.M. Zhang / Atmospheric Environment 109 (2015) 262e271 271

Pepermans, G., Driesen, J., Haeseldonckx, D., Belmans, R., D’haeseleer, W., 2005.Distributed generation: definition, benefits and issues. Energy Policy 33 (6),787e798.

Perry, S.G., 1992. CTDMPLUS: a dispersion model for sources near complex topog-raphy. Part I: technical formulations. J. Appl. Meteorol. 31 (7), 633e645.

Petersen, R.L., Cochran, B.C., Carter, J.J., 2002. Specifying exhaust and intake systems.ASHRAE J. 44 (8), 30e32.

Petroff, A., Zhang, L., 2010. Development and validation of a size-resolved particledry deposition scheme for application in aerosol transport models. Geosci.Model Dev. 3 (2), 753e769.

Porter, C.J., 2006. Determining stack height to prevent contamiant re-entry throughbuidling intakes. In: American Industrial Hygiene Association Conference.

Robins, A.G., Castro, I.P., 1977. A wind tunnel investigation of plume dispersion inthe vicinity of a surface mounted cubedII. The concentration field. Atmos.Environ. 11 (4), 299e311.

Shah, S.D., Cocker Iii, D.R., Johnson, K.C., Lee, J.M., Soriano, B.L., Wayne Miller, J.,2006. Emissions of regulated pollutants from in-use diesel back-up generators.Atmos. Environ. 40 (22), 4199e4209.

Sini, J.-F., Anquetin, S., Mestayer, P.G., 1996. Pollutant dispersion and thermal effectsin urban street canyons. Atmos. Environ. 30 (15), 2659e2677.

Steffens, J.T., Wang, Y.J., Zhang, K.M., 2012. Exploration of effects of a vegetationbarrier on particle size distributions in a near-road environment. Atmos. En-viron. 50, 120e128.

Steffens, J.T., Heist, D.K., Perry, S.G., Zhang, K.M., 2013. Modeling the effects of a solidbarrier on pollutant dispersion under various atmospheric stability conditions.Atmos. Environ. 69, 76e85.

Steffens, J.T., Heist, D.K., Perry, S.G., Isakov, V., Baldauf, R.W., Zhang, K.M., 2014.Effects of roadway configurations on near-road air quality and the implicationson roadway designs. Atmos. Environ. 94, 74e85.

Strachan, N., Farrell, A., 2006. Emissions from distributed vs. centralized generation:the importance of system performance. Energy Policy 34 (17), 2677e2689.

Tamura, T., 2008. Towards practical use of LES in wind engineering. J. Wind Eng. Ind.Aerodyn. 96 (10e11), 1451e1471.

Thompson, R.S., 1993. Building amplification factors for sources near buildings: awind-tunnel study. Atmos. Environ. Part A. General Top. 27 (15), 2313e2325.

Tominaga, Y., Mochida, A., Yoshie, R., Kataoka, H., Nozu, T., Yoshikawa, M.,Shirasawa, T., 2008. AIJ guidelines for practical applications of CFD to pedestrianwind environment around buildings. J. Wind Eng. Ind. Aerodyn. 96 (10e11),1749e1761.

Tong, Z., Wang, Y.J., Patel, M., Kinney, P., Chrillrud, S., Zhang, K.M., 2012. Modelingspatial variations of black carbon particles in an urban highway-buildingenvironment. Environ. Sci. Technol. 46 (1), 312e319.

Tseng, Y.-H., Meneveau, C., Parlange, M.B., 2006. Modeling flow around bluff bodiesand predicting urban dispersion using large eddy simulation. Environ. Sci.Technol. 40 (8), 2653e2662.

USEPA, 2004. User's Guide for the AERMOD Meteorological Preprocessor (AERMET).U.S. Environmental Protection Agency.

USEPA, 2004. Summary and Analysis of Comments: Control of Emissions fromNonroad Diesel Engines. U.S. Environmental Protection Agency.

USEPA, 2006. AP-42, Compilation of Air Pollution Emission Factors. U.S. Environ-mental Protection Agency.

USEPA, 2008. AERSURFACE User's Guide. U.S. Environmental Protection Agency.USEPA, 2014. Nonroad Compression-ignition Engineseexhaust Emission Standards.

U.S. Environmental Protection Agency.Wang, Y.J., Zhang, K.M., 2009. Modeling near-road air quality using a computational

fluid dynamics model. Environ. Sci. Technol. 43 (20), 7778e7783.Wang, Y.J., Zhang, K.M., 2012. Coupled turbulence and aerosol dynamics modeling

of vehicle exhaust plumes using the CTAG model. Atmos. Environ. 59, 284e293.Wang, Y.J., DenBleyker, A., McDonald-Buller, E., Allen, D., Zhang, K.M., 2011.

Modeling the chemical evolution of nitrogen oxides near roadways. Atmos.Environ. 45 (1), 43e52.

Wang, Y.J., Yang, B., Lipsky, E.M., Robinson, A.L., Zhang, K.M., 2013. Analyses ofturbulent flow Fields and aerosol dynamics of diesel engine exhaust inside twodilution sampling tunnels using the CTAG model. Environ. Sci. Technol. 47 (2),889e898.

Wang, Y.J., Nguyen, M.T., Steffens, J.T., Tong, Z., Wang, Y., Hopke, P.K., Zhang, K.M.,2013. Modeling multi-scale aerosol dynamics and micro-environmental airquality near a large highway intersection using the CTAG model. Sci. TotalEnviron. 443, 375e386.

Wood, C.R., Jarvl, L., 2012. Urban climate observations in Helsinki. Magzine Finn. AirPollut. Soc. (3), 30e33.

Wroblewska, M., 2011. Emergency generators- 10 second starting. ePower News (1).Xie, Z., Castro, I., 2006. LES and RANS for turbulent flow over arrays of Wall-

Mounted obstacles. Flow. Turbul. Combust. 76 (3), 291e312.Xie, Z.-T., Hayden, P., Wood, C.R., 2013. Large-eddy simulation of approaching-flow

stratification on dispersion over arrays of buildings. Atmos. Environ. 71, 64e74.Zhang, X., Zhang, K.M., 2015. Demand response, behind-the-meter generation and

ozone air quality. Environ. Sci. Technol. 49 (3), 1260e1267.