A framework for emissions source apportionment in industrial areas: MM5/CALPUFF in a near-field application

Abstract

This paper examines the relative source contribution to ground-level concentrations of carbon monoxide (CO), nitrogen dioxide (NO₂), and PM₁₀ (particulate matter with an aerodynamic diameter <10 μm) in a coastal urban area due to emissions from an industrial complex with multiple stacks, quarrying activities, and a nearby highway. For this purpose, an inventory of CO, oxide of nitrogen (NO_x), and PM₁₀ emissions was coupled with the non-steady-state Mesoscale Model 5/California Puff Dispersion Modeling system to simulate individual source contributions under several spatial and temporal scales. As the contribution of a particular source to ground-level concentrations can be evaluated by simulating this single-source emissions or otherwise total emissions except that source, a set of emission sensitivity simulations was designed to examine if CALPUFF maintains a linear relationship between emission rates and predicted concentrations in cases where emitted plumes overlap and chemical transformations are simulated. Source apportionment revealed that ground-level releases (i.e., highway and quarries) extended over large areas dominated the contribution to exposure levels over elevated point sources, despite the fact that cumulative emissions from point sources are higher. Sensitivity analysis indicated that chemical transformations of NO_x are insignificant, possibly due to short-range plume transport, with CALPUFF exhibiting a linear response to changes in emission rate. The current paper points to the significance of ground-level emissions in contributing to urban air pollution exposure and questions the viability of the prevailing paradigm of point-source emission reduction, especially that the incremental improvement in air quality associated with this common abatement strategy may not accomplish the desirable benefit in terms of lower exposure with costly emissions capping.

Implications:

The application of atmospheric dispersion models for source apportionment helps in identifying major contributors to regional air pollution. In industrial urban areas where multiple sources with different geometry contribute to emissions, ground-level releases extended over large areas such as roads and quarries often dominate the contribution to ground-level air pollution. Industrial emissions released at elevated stack heights may experience significant dilution, resulting in minor contribution to exposure at ground level. In such contexts, emission reduction, which is invariably the abatement strategy targeting industries at a significant investment in control equipment or process change, may result in minimal return on investment in terms of improvement in air quality at sensitive receptors.

Introduction

Air quality management (AQM) is an integrated approach necessitated by the multidimensional impacts of air pollution on public health, the physical environment, and community welfare (Longhurst et al., 2001; Cheung et al., 2005; World Health Organization [WHO], 2006). Despite economic and social considerations, scientific evidence of the contribution of a particular emission source to deteriorated air quality often constitutes the fundamental basis for regulatory decision-making, especially in contexts where several sources with different characteristics contribute to ambient air pollution. As such, establishing adequate spatial and temporal source–receptor relationships conveys a better understanding on what sources need to be targeted for mitigation measures (Laupsa et al., 2009). Although the latter is particularly important from a risk assessment and environmental justice perspective, the fact that higher source emissions is not necessarily associated with higher contribution to ground-level concentrations has driven regulatory agencies to establish standards on ambient concentrations in addition to those on emissions (Leksmono et al., 2006; Behera et al., 2011). For instance, industrial emissions occurring at high release heights experience greater atmospheric dilution capacity than ground-level releases, and thus air quality may not respond any better to costly point-source emission reduction. From a public policy perspective, although the latter has invariably been the obvious abatement strategy targeting industrial activity at a significant investment in control equipment or process change, ground-level area sources require equally closer examination, especially that these are becoming a common occurrence in urban regions, are often close to populated centers, and experience weak atmospheric dispersion leading to elevated exposure levels.

Nevertheless, industrial emissions in urban areas continue to be under scientific scrutiny, given their association with aggravated air quality in these areas (Krishna et al., 2005; El-Fadel et al., 2009; El-Fadel and Abi Esber, 2012), and their relative contribution to ambient air pollution assists decision-makers in implementing cost-effective management plans (Wang et al., 2005). However, working from measured pollutant concentrations at a particular receptor back to emissions requires an understanding of those processes that dictate plume evolution and transport, namely, the release height, meteorological factors, and chemical transformations. As such, source apportionment is not a straightforward activity and several approaches have been adopted to study the relative contribution of emission sources to ambient air pollution. Receptor modeling can be relied upon in such cases (Watson and Chow, 2004) using inverse dispersion modeling, source profiling, principal component analysis (PCA), and chemical mass balance (CMB) approaches. The latter can be resource demanding, are spatially limited to the studied receptor (Laupsa et al., 2009), and exhibit variable degrees of uncertainty (Hueglin et al., 2000). On the other hand, emission inventories in conjunction with atmospheric dispersion models are also used in this context (Islam, 1999; Crabbe et al., 2000; Fushimi et al., 2005), with the advantage of addressing the critical limitation of low spatial resolution that characterizes air quality monitoring at discrete receptors.

In this context, the current study aims at analyzing source apportionment to ambient air quality in a coastal urban area with emissions from an industrial complex, quarrying activities, and a nearby highway. Emission sources were broadly categorized as elevated point sources (i.e., industries) and ground-level area sources (i.e., highway and quarries) to examine the contribution of each source group to local air pollution in relation to precursor emissions and source characteristics, which therefore reflects on the efficacy of point-source emission control as the most common abatement strategy in mitigating air pollution. For this purpose, an inventory of carbon monoxide (CO), oxide of nitrogen (NO_x), and PM₁₀ (particulate matter with an aerodynamic diameter <10 μm) emissions was developed using the U.S. Environment Protection Agency (EPA) EPA-AP42 and European Environment Agency (EEA) emission factor guidelines (EPA, 1994; EEA, 2009). The resulting source emissions were tested for their representativeness by statistical analysis of CALPUFF simulations against year-round field measurements. The emission inventory was then coupled with the non-steady-state Mesoscale Model 5 (MM5)/CALPUFF model to simulate the dispersion of individual source emissions at several spatial and temporal scales. In order to establish the equivalence of two methods in estimating source contribution (namely, “emissions-in” and “emissions-out”), CALPUFF was tested for its sensitivity to input emissions and linearity in responding to changes in emission rates. A sensitivity analysis of CALPUFF simulations to chemical transformations was also conducted. Note that the linearity of CALPUFF in responding to input emissions has not been examined to date, particularly in contexts of short-range regulatory assessments where source groups have different geometry and are spatially stretched leading to rapid plume overlapping.

Methodology

Study area

The study area lies along the eastern Mediterranean (latitude 34˚27′N and longitude 35˚34′E), at around 60 km to the north of Beirut City (Figure 1a). The area is urban/semiurban to the west, and rural to the east, with a relatively complex terrain setting (Figure 1b). The general climate is of the subtropical, Mediterranean type with warm and dry summers and falls, and moderately cold and wet winters. The area hosts major cement and fertilizer industries clustered in and around the coastal town of Chekka. The characteristics of the industrial plants are summarized in Table 1, including type of activity, production rate, energy source, and reported control measures. Other significant emission sources in the area include vehicular emissions (highway) and quarrying sites. The spatial distribution of the emission sources over the study area, their different geometry, and distance to receptors (monitoring sites) (Figure 1a) are likely to lead to complex source–receptor relationships and plume mixing. Table 2 summarizes the physical characteristics of these sources relative to each other, which reveals that ground-level sources occupy considerable areas as opposed to spatially confined point-source stacks.

Table 1. Characteristics of the main plants in the study area ^a

	Production				Process Fuel		Energy Fuel
Industry Name	Process	Product (S)	Emission Control ^b	Rate (T/D)	Type	Quantity (T/D)	Type	Quantity (T/D)
Holcim (1 and 2)	Dry/preheater/precalciner	Clinker	ESP, BHF, C	5,800	Petcoke	580	Fuel oil	77
		Gray cement		3,190
Holcim (3)	Dry/preheater	Clinker	ESP, BHF, C	300	Petcoke	39	—	—
		White cement		167	Fuel oil	47
Cimenterie Nationale (CN)	Dry/preheater/precalciner	Clinker	ESP, BHF, C	2,900	Petcoke	480	Fuel oil	140
		2,000			Fuel oil	54
		Gray cement		1,380
Seament (CMO)	Dry/preheater	Clinker	ESP, BHF, C	200	Fuel oil	22	Fuel oil	1.7
		Gray cement		52
Lime & Plaster Co. (L&P)	Rotary kiln/ore/calciner	Lime	BHF	8.2	Fuel oil	2.7	—	—
		Plaster		1.8
Mineral Oil Recycling Co. (MORC)	Traditional batch	Base/lube oil		5.6	Gas oil	0.21	—	—
Lebanon Chemicals Co. (LCC)	Double contact/Rhone-Poulenc	Sulphuric acid	WS	650	Fuel oil	33	—	—
		Phosphoric acid		200
		TSP fertilizer		183
		SSP fertilizer		83
Notes: ^aSource: El-Fadel et al. (2009). ^bESP = electrostatic precipitator; BHF = baghouse filter; C = cyclone; WS = wet scrubber.

Table 2. Physical characteristics of modeled source groups

	Parameters
Industry Name	Stack Height (m)	Stack Diameter (m)	Gas Exit Temperature (K)	Gas Exit Velocity (m/sec)
Holcim (1 & 2)	108.5	3	393	12
Holcim (3)	51	2.1	403	12
Cimenterie Nationale (CN)	75	4	393	14
Seament (CMO)	55	1	393	12
Lime & Plaster (L&P)	52	0.75	398	12
Mineral Oil Recycling Co. (MORC)	10	0.3	433	6
Lebanon Chemicals Co. (LCC)	38	0.9	353	9
Area Sources	Dimensions (m^2)
Highway	Series of rectangular area sources with total area of 16,000 m × 16 m
CN quarry	Polygonal area source with an area of 510,500 m^2
Holcim quarry	Polygonal area source with an area of 93,705 m^2
LP quarry	Polygonal area source with an area of 151,800 m^2

Figure 1. Characteristics of the study area (colour figure avilable in online).

Field measurements

Figure 1 shows the location of the monitoring sites at five towns, namely, Enfeh (LO1), Chekka (LO2), Fih (LO3), Kfarhazir (LO4), and Kefraya (LO5). The terrain elevations show that LO1 and LO2 lie on a flat terrain close to the coastline, whereas LO3, LO4, and LO5 are located in hilly areas at 200 to 400 m above sea level. Ambient concentrations of CO and nitrogen dioxide (NO₂) were monitored using an environmental monitoring station (EMS). PM₁₀ readings were recorded using an SKC Split 2 real-time dust monitor. Daily average concentrations of air quality indicators were computed using 30-min or 10-min instantaneous EMS readings and 10-s and 1–5-min average PM₁₀ readings. The field data set was examined for discrepancies that could be attributed to measurement errors. For instance, although CO and NO₂ were monitored simultaneously, some NO₂ records showing repetitive trends where measurements infrequently meander around a constant value were excluded from the analysis. PM₁₀ measurements were conducted at discrete days during each period. Table 3 summarizes the monitoring periods at each location along with the total number of daily average records available for statistical analysis upon quality assurance. The periods cover 10 months including the two dominant meteorological regimes of the study area, namely, the warm and dry season that usually spans from May to October, and the wet season that spans from November to April.

Table 3. Summary of field monitoring campaign for CO, NO₂, and PM_{1 0}

			Number of Daily Records ^b
Designation	Location	Monitoring Periods ^a	CO	NO2	PM10 ^c
LO1	Enfeh	Jan 01–Jan 11; Apr 19–May 13; Sep 09– Oct 05	59	48	6
LO2	Chekka	Jan 11–Feb 02; May 13–May 20; Sep 06–Sep 23	42	20	3
LO3	Fih	Feb 03–Feb 19; Jun 15–Jul 05; Aug 26–Sep 06	45	35	3
LO4	Kfar-hazir	Mar 01–Mar 15; Jul 05–Aug 16	40	40	3
LO5	Kefraya	Mar 15–Mar 23; Jun 15–Jul 05	28	8	3
Notes: ^aAll measurements were conducted during the year 2003. ^bTotal number of days considered for statistical analysis. ^cPM10 measurements were conducted at selected discrete (nonconsecutive) days at each location.

Model description

The MM5/CALMET/CALPUFF modeling system was used to simulate the dispersion of CO, NO_x, and PM₁₀ from various sources. The system is recommended by the EPA for long-range transport regulatory modeling and, on a case-by-case basis for near-field applications where non-steady-state assumptions are significant (EPA, 2008). It includes a meteorological processor (CALMET), a dispersion model (CALPUFF), and a post-processor (CALPOST). CALMET is a diagnostic wind field model with micrometeorological modules for overwater and overland boundary layers (Scire et al., 2000). CALPUFF is a multilayer, multispecies non-steady-state Gaussian puff dispersion model that simulates the effects of time- and space-varying meteorological conditions on pollutant transport, transformation, and removal. Unlike traditional Gaussian plume models, CALPUFF simulates the continuous plume from a source as a series of discrete “puffs” (packets of pollutants) (Scire et al., 2000) that are transported and dispersed through a three-dimensional wind and micrometeorological field. The CALPUFF system has been applied in contexts of exposure assessments (Macintosh et al., 2010), health impacts (Levy et al., 2002), and source apportionment studies (Yim et al., 2010). This study couples the Mesoscale Model 5 (MM5) with CALMET to drive the CALPUFF simulations. Four-kilometer-resolution MM5 data were initially used to generate large-scale wind fields, which were adjusted afterwards using CALMET to reflect the high-resolution terrain and land use data in the area. The meteorological and computational grids were 50 × 50 km with a resolution of 0.5 × 0.5 km. Dispersion options were set to model defaults.

Emission factors

In the absence of representative local emission factors (EFs), ranges of internationally reported values were considered and adjusted to account for case-specific conditions (quality of raw material, processing, and fuel types, etc.) (El-Fadel and Abi-Esber, 2012; Whyatt et al., 2007). Emission rates (ERs) (mass of pollutant per unit time) were then derived based on the EFs (mass of pollutant per mass or volume of product) and the activity rate (AR) (mass of product per unit time), taking into consideration the percent reduction (R) associated with the presence of control equipment.

Table 4 summarizes the ranges of emission rates (g/sec) derived for the pollutants (CO, NO_x, and PM₁₀) (El-Fadel and Abi-Esber, 2012). The range of the EPA emission factors (EPA, 1994) was determined depending on the performance of control devices, fuel composition, and source operational characteristics. As such, efficiencies of PM₁₀ control devices were considered to range between 90% and 99% for electrostatic precipitators, 95% and 99% for bag house filters, 75% and 95% for cyclones, and 85% and 97% for wet scrubbers. In the context of emissions from fuel combustion, nitrogen contents ranging from 0.5% to 0.6% (the used range of nitrogen mass content was based on values collected during field interviews with the industries) were assumed for residual oil. CO emission factors were assumed to vary by up to a factor of 10. Average vehicle speeds ranging between 55 and 60 miles per hour were used to determine highway emissions.

Table 4. Range of emission factors (g/sec)

	CO		NOx		PM10
Emittant Source (see Figure 1)	EEA(Min–Avg–Max)	USEPA(Min–Max)	EEA(Min–Avg–Max)	USEPA(Min–Max)	EEA(Min–Avg–Max)	USEPA(Min–Max)
Point sources
Holcim (1 & 2)	104.1–208.1–416.2	15.5–46.5	41.6–161.3–624.3	120.6–212.6	0.5–4.4–12.8	326.5–825.7
Holcim (3)	5.4–10.8–21.6	0.43–5.03	2.2–8.4–32.4	8.4–13.8	0.055–0.83–2.16	4.7–49.9
Cimenterie  Nationale (CN)	72.7–145.4–290.7	26.44–52.9	29.1–112.7–436.1	152–229.6	0.607–9.54–27.9	494.9–1861.3
Seament (CMO)	2.9–5.8–11.7	0.4–1.4	1.2–4.5–17.5	2–6.6	0.001–0.007–0.03	12.9–31.9
Lime & Plaster Co.  (L&P)	0.03–0.20–1.40	0.014–0.16	0.05–0.30–1.40	0.21–0.43	0.028–0.25–0.57	1–2
Mineral Oil Recycling Co.  (MORC)	0.001–0.0017–0.002	0.001–0.01	0.002–0.005–0.01	0.05–0.05	0.000007–0.0004–0.0001	0.0011–0.0011
Lebanon Chemicals  Co. (LCC)	0.3–0.7–10	0.2–2.3	0.09–1.7–2.6	3.4–3.8	0.012–0.105–0.3	1.6–2.2
Area sources
Holcim quarry	—	—	—	—	0.095–0.57–2.84	30.9–36.6
CN quarry	—	—	—	—	0.051–0.36–2.06	112.3–142.9
L&P quarry	—	—	—	—	0.00012–0.00061–0.00307	16.7–16.7
Line sources
Highway	3.55–77.27–154.54	2472– 4172	0.31–7.73–13.91	618.15–726.33	0.0062–0.011–0.012	—

On the other hand, the ranges of EEA emission factors (EEA, 2009) were calculated based on Tier 1 and 2 emission factors and corresponding 95% confidence intervals. The emission factors, which are reported in g/ton of product, were multiplied by the size of production of individual industries (eq 1) to yield a g/sec total emission rate. In cases where emission factors are reported in g/GJ of energy consumed, heat values of 29.3, 45.0, and 32.6 GJ/ton of gas oil, fuel oil, and petcoke, respectively, were used along with the industries' total process fuel consumption to compute emission rates. Whenever Tier 2 emission factors were used, emissions from storage, handling, and transport of product were calculated separately and added to the final emission factor (this type of emissions is implicitly accounted for in Tier 1 emission factors). In the case of cement, lime, and plaster production, 99%, 95%, and 90% reduction in minimum, average, and maximum PM₁₀ emission factors were considered because the EEA emission factors assume that only electrostatic precipitator (ESP) control is in place, whereas field interviews with the industries showed that bag house filters are also being used (El-Fadel et al., 2009).

(1)

Statistical analysis

A total of 13 monitoring periods (see Table 3) coupled with five emission factors (EEA_min, EEA_avg, EEA_max, EPA_min, EPA_max) resulted in 65 simulations (13 simulations at each emission factor) and provided sufficient spatial and temporal variability to evaluate the representativeness of each emission factor. As such, model performance at each emission factor was evaluated using several statistical indicators (fractional bias [FB], geometric mean bias [MG], normalized mean square error [NMSE], and the fraction of predictions within a factor of 2 of observations [FAC2]) and data plots correlating model predictions with corresponding field measurements (Chang and Hanna, 2004). FB is a linear measure, and reflects the degree of matching between the predicted and observed mean of the concentration distribution. It ranges between −2 and +2, with a perfect model resulting in FB = 0. MG is also linear, and reflects the degree of bias of the geometric mean rather than the arithmetic mean. A perfect model would result in MG = 1. NMSE measures the relative scatter of the concentration distribution and a perfect model would result in an NMSE = 0. FAC2 is the most robust measure, as it reflects the percentage of predicted concentrations lying within a factor of 2 of observations. Scatter and Q-Q plots were also used to provide visual insight into model's performance. Scatter plots are intended to reveal the scatter about the 1:1 (y = x) line by plotting predicted (y-axis) against the corresponding observed (x-axis) concentrations, which indicates model tendency towards under/overpredictions. Q-Q plots start with the same set of predicted and observed concentrations, but rather ranks these from smallest to greatest so that the 1st, 2nd, …, nth highest predicted concentrations would be plotted against the corresponding 1st, 2nd, …, nth highest observed concentrations, revealing thus model biases in matching the upper end of the concentration distribution. The latter is of particular interest in regulatory contexts of air quality modeling.

Source apportionment

At the estimated emission factors, source apportionment was conducted at various receptors to assess the contribution of each source category to ambient concentrations of air quality indicators at several locations and the likelihood of under/overestimations of source emissions. In the interest of comparative assessment, the sources were categorized as point (Industrial facilities in Table 1) and area (highway and quarrying sites) sources. Source apportionment was conducted by designing a set of emission sensitivity simulations to test the linear response of the model to gross changes in the emission rate (Barna et al., 2006). With the “base-case simulations” at hand (simulations where all source categories were included and considered to be emitting), the contribution of a particular source group to the pollutant concentration at a certain receptor was evaluated in two ways:

1.	“Emissions-in” approach: This scenario for a simulation sets the emission rate of all sources to zero, except for the source to be assessed. The contribution of this particular source to the receptor is the predicted concentration.
2.	“Emissions-out” approach: This scenario is the same as the “base case” (all sources emitting), but removes only the emissions from the source to be assessed. The source contribution is thus the difference between the predicted concentration in the “base case” and the predicted concentration in the “emissions-out” simulation.

This process is intended to test the linearity of the model in responding to changes in precursor emissions, and provides an insight on its response to rescaling of emission factors. In essence, for a given source, the model is assumed to be linear if the predicted concentration contributed by this source is the same when simulated via the two scenarios. In other words, the sum of the concentration predicted in the “emissions-in” and “emissions-out” simulations has to be equal to that predicted in the “base case.” However, this might not be valid for reactive pollutants, which are subject to chemical transformations in the atmosphere that are dependent on their relative ambient mass available for chemical reactions. For this purpose, uncertainty in the results of CALPUFF response to changes in emission rates is also discussed in the context of sensitivity analysis to chemical transformations of NO_x to nitrate aerosols. The chemistry model MESOPUFF II scheme within CALPUFF utilizes first-order reaction kinetics for these transformations, and computes daytime and nighttime conversion rates for chemical reactions. The NO_x conversion rate is proportional to the ambient ozone concentration and plume NO_x concentration, and inversely proportional to a stability index (ranging between 2 and 6 based on five PG stability classes). Due to the lack of ambient ozone concentrations in the study area, model defaults of ozone and ammonia background concentrations (80 and 10 ppb, respectively) were used, and were considered to be uniform throughout the year. As for nighttime conversion rates, model defaults of NO_x conversion rate at 2% per hour was used.

The contribution of source groups to the total concentration was evaluated using long-time averages and time series of the 24-hr averages to provide a contribution assessment over finer temporal scales. In a sampling period of more than 1 week, the peak (maximum) week-average concentration was considered. The percent contribution of each source group to this concentration was then calculated to evaluate source contribution at peak levels. Furthermore, one typical time period at each location was analyzed, namely, those where the model exhibited relatively better performance, as these are considered the most reliable for such analysis.

Results and Discussion

Estimation of emission factors

For the purpose of brevity, the current section presents statistical analysis and model performance evaluation at emission factors that showed best correlations with observational data. For instance, CALPUFF simulations using the emission factors EEA_min, EEA_avg, and EPA_min exhibited weak agreement with field measurements at all locations and for all indicators. In contrast, simulations using EEA_max exhibited an acceptable match with CO observations most frequently, whereas those with EPA_max matched better with NO₂ and PM₁₀ observations. Table 5 summarizes the statistical analysis of CALPUFF predictions against observations for CO (using EEA_max emission factor) and NO₂ (using EPA_max emission factor). Except at LO3 (Fih), the statistical parameters indicate satisfactory to good agreement between simulated and observed CO concentrations. At LO1 (Enfeh), LO2 (Chekka), and LO5 (Kefraya), the FB ranges between 0.04 and 1.2 with an average of 0.24, indicating excellent to acceptable matching of the mean concentration. The average FAC2 value at these locations is 48%, suggesting a reasonable match between predicted and observed concentrations, particularly at LO5 (Kefraya) and LO2 (Chekka) where predictions lie within a factor of 2 of observations for 75% and 67% of the times, respectively. MG and NMSE values further indicate that CALPUFF simulations reproduced measured CO concentrations fairly well, with best MG of 0.97 (perfect value is 1) occurring at LO1 (Enfeh) and best NMSE of 0.3 (perfect value is 0) occurring at LO2 (Chekka). Despite weaker correlations, statistical analysis at LO4 (Kfarhazir) reflects acceptable agreement between CO predictions and corresponding observations, with an overall tendency towards underprediction. NO₂ predictions reveal a similar trend, with best matching of observations at LO1, LO2, and LO5, which are the closest locations to the sources (Figure 1), particularly highway emissions. Simulated NO₂ concentrations at LO3 and LO4 show weaker agreement with observations, with a tendency towards underpredictions.

Table 5. Statistical analysis of model simulations against field measurements for CO and NO₂

Pollutant		CO				NO2
Location	Period	FB	MG	NMSE	FAC2(%)	FB	MG	NMSE	FAC2(%)
Enfeh (LO1)	Jan 01–Jan 11	0.57	2	0.4	50	—	—	—	—
	Apr 19–May 13	1.2	5	3	10	0.38	0.8	1.9	25
	Sep 09–Oct 05	−0.5	0.97	2.3	31	0.8	3.3	4.9	15
	Aggregate^a	0.6	2.1	2	28	0.78	2	4.9	19
Chekka (LO2)	Jan 11–Feb 02	0.48	2.1	0.6	55	0.54	1	1.9	35
	May 13–May 20	0.04	1.2	0.3	67	—	—	—	—
	Sep 06–Sep 23	1.2	6.7	3	25	—	—	—	—
	Aggregate	0.8	3	1.4	41	—	—	—	—
Fih (LO3)	Feb 03–Feb 19	1.8	405	32	3	1.8	52.2	22.8	10
	Jun 15–Jul 05	1.7	28	16	2	1.97	223	225	2
	Aug 26–Sep 06	1.1	4.6	2.4	40	—	—	—	—
	Aggregate	1.7	50	14.1	10	1.9	223	285	4
Kfarhazir (LO4)	Mar 01–Mar 15	0.78	3	1.8	24	1.1	6	2.4	29
	Jul 05–Aug 16	1.4	19.7	5.8	18	1.4	4.5	12	17
	Aggregate	1.2	9.7	4	15	1.4	5.1	8.4	16
Kefraya (LO5)	Mar 15–Mar 23	−0.36	0.8	0.6	75	−0.54	0.4	1.3	38
	Jun 15–Jul 05	−0.7	0.6	3.2	70	—	—	—	—
	Aggregate	−0.72	0.64	3.5	71	—	—	—	—
Notes: ^aAggregate: Total observations at a receptor were compared with predictions regardless of time period.

Overall, noticeable matching between predicted and observed CO and NO₂ concentrations occurred at LO1, LO2, and LO5 where the source groups considered in model simulations may dominate the contribution to pollutant concentrations, particularly that these are the closest receptors to highway and point-source emissions. On the other hand, weaker correlations occurring at the farther locations (LO4 and LO5) can be attributed to the fact that other unaccounted for sources (domestic generators used during electricity blackouts, arterial roads, open burning of residual waste, etc.) may contribute to significant background concentrations, or otherwise to errors in simulated meteorology. Similarly, Table 6 summarizes a similar analysis for PM₁₀ predictions. Note that the measured daily average PM₁₀ concentrations were aggregated at each location because these occurred during discrete nonconsecutive days, and then compared with corresponding predictions. Excellent agreement is noticeable at all locations, with FAC2 reaching 100% at LO3, LO4, and LO5, which are the closest locations to quarries (Figure 1). The FB, MG, and NMSE values at all locations reveal very good model performance in reproducing PM₁₀ measurements at all locations.

Table 6. Statistical analysis of model simulations against field measurements for PM₁₀

		PM10
Designation	Location	FB	MG	NMSE	FAC2(%)
LO1	Enfeh	0.06	1	0.83	50
LO2	Chekka	0.8	1.95	1.4	67
LO3	Fih	−0.1	0.98	0.09	100
LO4	Kfar-hazir	−0.21	0.85	0.17	100
LO5	Kefraya	−0.56	0.56	0.33	100

Figure 2 shows typical scatter and Q-Q plots for CO, NO₂, and PM₁₀ at selected locations. All scatter plots reveal good model performance, as most predicted concentrations lie within a factor of 2 of observations. Q-Q plots indicate good matching between predictions and observations for the three indicators, particularly in the ability of predicting the peak concentrations, despite the slight underprediction of the peak NO₂ concentration. Overall, both scatter and Q-Q plots reveal a tendency towards underprediction of observed concentrations for all indicators (most data points lie below the 1:1 line), but always within acceptable differences (factor of 2 underprediction). The estimated emission factors for the three indicators are presented in Figure 3. Industrial facilities (point sources) accounted for 81%, 37%, and 93% of total CO, NO_x, and PM₁₀ emissions, respectively. Highway emissions accounted for the remaining 19% of CO and 67% NO_x emissions, and quarries for the remaining 7% PM₁₀ emissions. However, the spatial distribution of sources along with their different geometry (dispersion characteristics) does not allow for correlation between emission magnitudes and contribution to ambient pollutant concentrations.

Figure 2. Selected (a) scatter and (b) Q-Q plots for CO (µg/m³), NO₂ (µg/m³), and PM₁₀ (µg/m³).

Figure 3. Emission characteristics in the study area.

Emission sensitivity

The contribution of highway emissions to CO and NO₂ concentrations was calculated through the “emissions-in” (EI), and the “emissions-out” (EO) simulations. To test the linearity of CALPUFF, the ratio of predicted 24-hr average concentrations in the EI simulation to that in the EO simulation is shown in Figure 4a–d, at various receptors and time periods. Clearly, highway contribution to CO and NO₂ levels is equivalent in the two scenarios, with a ratio of EI/EO always unity, except for 1 day at LO2 (Figure 4a) and 3 days at LO1 (Figure 4c), but with negligible deviation (0.9–1.2). It is worth noting that those days of overprediction of the highway contribution (EI/EO > 1) were characterized by higher wind speeds and less calms frequency, as opposed to days of underpredictions. For instance, the day with underprediction at LO2 was characterized by high percentage of calm hours (11 out of 24) and low average wind speed (0.6 m/sec), which may explain the underestimation of the highway contribution (EI/EO = 0.9) during that day. Days of overpredictions, however, were characterized by zero calm hours and high wind speeds (4.6 m/sec), and CALPUFF overestimated the contribution of the highway. Given the equivalence between the “emissions-in” and the “emissions-out” scenarios, the remaining discussion on source contributions considers the “emissions-out” simulations only.

Figure 4. Ratio of the 24-hr average concentration predicted via the EI and EO scenarios.

Carbon monoxide apportionment

Table 7 shows the contribution of highway and stack emissions to the maximum weekly average CO concentration at the five locations. At the closest locations LO2 and LO5, highway emissions contributed 96% and 99% of the predicted CO concentrations, respectively, which matched well with the corresponding observed levels. The two source groups contributed almost equally at the northerly and farthest receptors LO1 and LO3, with noticeable underprediction of the weekly average CO concentration at LO3. At LO4, highway emissions accounted for 77% of the predicted concentration, which was around a factor of 3 less than the observed level.

Table 7. Source contribution to the highest weekly average CO concentration

Location (Period)	Observed (µg/m^3)	Predicted (µg/m^3)	Point Sources (%)	Highway (%)	Quarries (%)
LO1 (Jan 01–Jan 11)	274	110	53	47	0
LO2 (Jan 12–Feb 02)	550	440	4	96	0
LO3 (Jun 15–Jul 05)	800	35	54	46	0
LO4 (Jul 06–Aug 10)	700	200	23	77	0
LO5 (Jun 15–Jul 05)	940	1780	1	99	0

Figure 5 shows time series of the daily contribution of highway and point-source emissions to the CO concentration at several locations. Simulated CO levels matched relatively well the observations at LO1 and LO2 (within a factor of 2) (Figure 5a–b), with underprediction when stack emissions dominated or contributed significantly to the CO concentration. For instance, underprediction can be noted during the first week in the simulated period at LO1 (Figure 5a) and the last week at LO2 (Figure 5b), where stack emissions contributed 65–80% and 40–60%, respectively. Furthermore, at LO3 (Figure 5c), CALPUFF underestimated the maximum 24-hr average CO concentration for the entire simulated period. Note that point sources accounted for around 85% of the predicted concentration during some days (June 25–26), and for 54% of the highest weekly average. On the other hand, predicted concentrations were relatively closer to observations when highway emissions dominated the contribution to CO levels, namely, the second half of the simulated period at LO3 where the highway accounted for 45–70% of predicted concentration. LO4 reveals a similar trend (Figure 5d), with underpredictions of the 24-hr average concentration when point sources dominated the contribution, namely, the 1st week of the simulation period (Figure 5d). However, highway emissions dominated the contribution to CO levels at this receptor, reaching up to 95% at some days, and where CALPUFF predictions matched well the magnitude and temporal pattern of observations (less than a factor of 2 underprediction). At LO5, highway emissions contributed 100% of the predicted concentration for all days, with overpredictions of the observed concentration, and thus was excluded from further analysis. In essence, underpredictions of the observed CO concentrations are noticeable at LO3 and LO4, which can be attributed to underestimation of CO emissions from point sources. Also, vehicular emissions are significant contributors to the CO concentration in the study area, with this contribution decreasing at farther locations.

Figure 5. Typical daily source contribution to CO concentration (µg/m³) at several locations.

To provide a contribution assessment over the entire area, Figure 6 shows contour lines delineating CO levels contributed by highway and point-source emissions separately, during 2 months of the year. The highest 24-hr average CO concentration contributed by highway emissions increased by around a factor of 3 during summer (3955 µg/m³ during January vs. 11,889 µg/m³ during July), but occurred at the same location, close to LO5. Also, summer times were characterized by higher near-source accumulations (Figure 6c), primarily due to the frequent occurrence of stagnation hours. For instance, LO1 and LO4 (relatively distant receptors from highway emissions) witnessed CO levels of 500 and 750 µg/m³ during winter (Figure 6a), respectively, as opposed to lower levels (250 µg/m³) during the summer (Figure 6c). Highway emissions were thus transported to farther locations during winter times, but always with concentrations lower than the ambient air quality standard. On the other hand, the peak 24-hr average CO concentration contributed by point sources decreased from 668 µg/m³ during winter to 522 µg/m³ during summer (Figure 6b–d). Furthermore, winter times were characterized by larger areas affected by higher CO exposure albeit lower than the standard.

Figure 6. Spatial distribution of the 24-hr average CO concentration during winter and summer (colour figure avilable in online).

Nitrogen dioxide apportionment

Table 8 summarizes the contribution percentages of highway and stack emissions to NO₂ concentration at the five locations, indicating the prevalence of highway emissions (ranging between 87% and 100%) on the simulated NO₂ concentration. Significant underprediction of the observed concentration is noticeable at LO1 and LO3. The contribution of the two categories to daily NO₂ concentration is shown in Figure 7a–d. The predicted 24-hr average concentration matched well observations at LO2, except for 2 days (January 16–17), when underprediction is evident (Figure 7a). Highway emissions accounted for around 95% of the NO₂ concentration at this receptor for all days. At LO4 (March 2–15) (Figure 7b), CALPUFF underpredicted the observed concentration (a factor of 3), but was able to reproduce the temporal trend over the time period. Highway contribution prevailed also at this receptor, with a percentage of >95%. At LO3, point sources contributed up to 36% during some days (Figure 7c), but with significant underpredictions of observations (2–3 orders of magnitude), which are hence not included. Although this could be attributed to underestimation of emissions from point sources, there remains the possibility that other potential unaccounted-for sources in the simulations (domestic heating/generators, arterial roads, open burning, etc.) are contributing to such elevated concentrations. Furthermore, CALPUFF generally captured the temporal fluctuation in the concentration distribution at this receptor (LO3), indicating that a higher emission rate would result in better matching of observed concentrations. During another period at LO4 (Figure 7d), the simulated concentration was 5 times less than the observed, with highway emissions dominating the contribution (90%), except for July 10–14, when point sources contributed ∼35%.

Table 8. Source contribution to the highest weekly average NO₂ concentration

Location (Period)	Observed (µg/m^3)	Predicted (µg/m^3)	Point Sources (%)	Highway (%)	Quarries (%)
LO1 (Jan 01–Jan 11)	95640	624	6	94	0
LO2 (Jan 12–Feb 02)	3010	1910	1	99	0
LO3 (Jun 15–Jul 05)	24550	100	13	87	0
LO4 (Jul 06–Aug 10)	1840	455	2	98	0
LO5 (Jun 15–Jul 05)	1670	3102	0	100	0

Figure 7. Typical daily source contribution to NO₂ concentration at several locations.

The spatial distribution of NO₂ levels contributed by the two categories is shown in Figure 8. Summer times were characterized by higher accumulation of NO₂ concentration. Peak concentrations contributed by point sources increased during summer and occurred at a different location (southwest in winter and north in summer) (Figure 8b–d). Also, larger areas were affected by point-source emissions during summer, with relatively higher concentrations. In contrast, highway emissions contributed to larger areas during the winter, indicating longer transport distances. The peak concentration occurred at the same location near LO5, and increased by slightly more than a factor of 2 in the summer.

Figure 8. Spatial distribution of the 24-hr average NO₂ concentration during winter and summer (colour figure avilable in online).

Particulate matter apportionment

Sources contributing to PM₁₀ emissions are area sources (quarrying sites) (197 g/sec) and point sources (2773 g/sec), accounting for 7% and 93% of total emissions (2784 g/sec), respectively. Measurements of ambient PM₁₀ concentrations covered 1–2 days at a receptor. The corresponding contribution of each source category to PM₁₀ levels was examined in the context of spatial analysis of the concentration distribution. The analysis considered one period of the year to reflect on the contribution magnitude of each source category, rather than the spatial distribution (areas affected), which is expected to have similar patterns as that of NO₂. Figure 9 shows contour lines of PM₁₀ levels in the study area contributed by point and area sources during summer times (July), which revealed high accumulation levels.

Figure 9. Spatial distribution of the 24-hr average PM₁₀ concentration during summer (colour figure avilable in online).

Emissions from point sources affected almost all the study area, with concentrations exceeding the 24-hr average standard, with inner areas witnessing higher PM₁₀ levels than coastal ones (Figure 9a). The peak 24-hr average concentration contributed by point sources reached 4186 µg/m³ south of LO4. On the other hand, emissions associated with quarrying (area sources) affected a smaller part of the study area, albeit higher accumulation levels near the sources with peak concentration reaching 17,700 µg/m³ close to the CN quarry, indicating weak dispersion and shorter transport.

Chemical transformations

Figure 10 shows time series of the 1-hr average NO₂ concentration at several locations simulated with and without considering chemical transformations. Although time series of both underpredicted observed NO₂ concentrations, particularly at LO3 (several orders of magnitude and hence not included), they coincided with each other, indicating little to no NO_x reactions. Only at LO3 (Figure 10c), simulated NO₂ concentrations decreased slightly when chemical transformations (CT) were incorporated, reflecting the significance of longer transport distances on inducing chemical reactions. Note that the statistical parameters calculated upon simulating chemical transformations, and the spatial distribution of the NO₂ concentration, exhibited no significant differences compared with base-case simulations.

Figure 10. Time series of NO₂ concentration with and without chemical transformations.

Conclusion

The non-steady-state CALPUFF model was coupled with an emission inventory for source apportionment to CO, NO₂, and PM₁₀ levels at several locations in an urban industrial area encompassing emissions from a multistack complex, a highway, and quarrying sites. CALPUFF was tested for its linear response to changes in emissions, and exhibited equivalence in simulating the contribution of a source category using two approaches, namely, the “emissions-in” and “emissions-out” scenarios, which indicates that the model adds the contributions from each source category to the pollutant concentration at a receptor. Source apportionment revealed that ground-level releases (i.e., highway and quarries) extended over large areas dominated the contribution to ambient concentrations of CO, NO₂, and PM₁₀ at close locations to the sources, despite the higher cumulative emissions from point sources, indicating that releases at elevated heights experience adequate dilution. Near the highway and quarries, significant accumulation is noticeable during summer times compared with winter times due to higher frequency of calm conditions and unfavorable meteorology, with point sources affecting distant areas the most, but with lower exposure levels compared with area sources. The latter results point to the significance of ground-level area sources in contributing to elevated exposure levels in urban areas and that such sources should be equally targeted for management options. Equally important, the results raise questions in relation to the viability of the prevailing paradigm of point-source emission reduction that encompass excessive costs with little return on investment measured as improvement in air quality or decreased exposure per unit cost invested. To this end, sensitivity analysis to chemical transformations, although limited by the use of model defaults of background ozone and ammonia concentrations, revealed that these were insignificant and subsequently had little effect on the linear response of CALPUFF to NO_x emissions, possibly due to short-range plume transport within the study area. Note that for industry-specific pollutants that constitute markers for certain industries and that are not commonly emitted from open area/ground-level sources in urban settings, point-source emission reduction remains the critical abatement measure, with source apportionment becoming of less significance.

Acknowledgment

The air quality and meteorology database was collected by the Lebanese American University (LAU) under a grant from the United States Agency for International Development (USAID). Special thanks are extended to Drs. Gebran Karam and Mazen Tabbara at LAU for making the database available.