Estimation of hydrogen sulfide emission rates at several wastewater treatment plants through experimental concentration measurements and dispersion modeling

Abstract

The management and operation of wastewater treatment plants (WWTP) usually involve the release into the atmosphere of malodorous substances with the potential to reduce the quality of life of people living nearby. In this type of facility, anaerobic degradation processes contribute to the generation of hydrogen sulfide (H₂S), often at quite high concentrations; thus, the presence of this chemical compound in the atmosphere can be a good indicator of the occurrence and intensity of the olfactory impact in a specific area. The present paper describes the experimental and modelling work being carried out by CEAM-UMH in the surroundings of several wastewater treatment plants located in the Valencia Autonomous Community (Spain). This work has permitted the estimation of H₂S emission rates at different WWTPs under different environmental and operating conditions. Our methodological approach for analyzing and describing the most relevant aspects of the olfactory impact consisted of several experimental campaigns involving intensive field measurements using passive samplers in the vicinity of several WWTPs, in combination with numerical simulation results from a diagnostic dispersion model. A meteorological tower at each WWTP provided the input values for the dispersion code, ensuring a good fit of the advective component and therefore more confidence in the modelled concentration field in response to environmental conditions. Then, comparisons between simulated and experimental H₂S concentrations yielded estimates of the global emission rate for this substance at several WWTPs at different time periods. The results obtained show a certain degree of temporal and spatial (between-plant) variability (possibly due to both operational and environmental conditions). Nevertheless, and more importantly, the results show a high degree of uniformity in the estimates, which consistently stay within the same order of magnitude.

Implications:

Estimating emissions to the atmosphere is usually considered a complex task, especially when such discharge comes from diffuse or uncontrolled sources. In any approach to air quality control, just from the point of view of increasing knowledge or as a management problem in order to reduce present levels of pollution, accurate estimation of emission rates is revealed as a fundamental step. Evaluation from an indirect method provides a useful methodology in such cases. Combination of dispersion modeling with experimental air concentration measurements permits one to obtain a first estimation of H₂S emission rates at several wastewater treatment plants. In a subsequent refinement of the process, the initial constant average emissions calculated were improved, leading to the formulation of a time-varying emission model, as a function of environmental quantities.

Introduction

All human settlements must deal with the problem of sewage. Although sewage treatment and disposal mainly affect the solid (sludge) and liquid phases, they can also impact the gaseous phase by releasing malodorous chemical compounds into the atmosphere (Thorkild and Vollertsen, 2001).

Previous work done at several wastewater treatment plants (WWTPs) (Coll-Lozano et al., 2005; Coll-Lozano et al., 2006) shows that the main sources responsible for causing odors associated with the release of hydrogen sulfide in WWTPs are:

•	Pretreatment of influent wastewater: wastewater pumping, operations of grinding and degritting, degreasing.
•	Sludge treatment: blending, stabilization, dewatering, ultimate disposal.

Other operations that can emit hydrogen sulfide (although in lesser amounts than those just listed) are those involved in:

•	Primary clarifiers (medium emission potential).
•	Biological clarifiers (medium/low emission potential).
•	Secondary clarifiers (low emission potential).

The olfactory impact can be analyzed from many different perspectives and methodologies (psychometry, field inspections, dynamic olfactometry, electronic noses, etc.). The method described in this paper, the use of passive samplers to measure the concentration levels of specific compounds associated with malodor, has certain advantages over other methodologies like olfactometry in that it can be used to monitor air quality over extensive areas with very good spatial resolution. Moreover, it is a low-cost, easily tuned, and relatively simple technology. The possibilities of passive sampling devices as environmental monitors have been widely described in the literature (Campbell et al., 1994; De Saeger et al., 1991; De Saeger et al., 1995; Hewitt, 1991) and there are many commercial products on the market for different chemical species (SO₂, NO₂, _O3, CO, benzene). Passive samplers are very well suited for use in systematic sampling campaigns: (1) to collect information on the temporal variability of the system; (2) to experimentally characterize the spatial distribution of the ground-level concentrations field on the basis of a large number of measuring points; (3) to establish cause–effect relationships between plant operations and experimental ground-level concentration, thus enabling mitigation strategies to be formulated; (4) to compare and contrast experimental data with numerical simulation results; and so on. One limitation of this technique is its poor temporal resolution. This can be effectively offset with the use of continuous measurement sources that provide information on temporal variability, as well as with the use of numerical models.

An interesting approach to the olfactory impact problem is to view it as a typical problem of pollutant dispersion (Llavador et al., 2010), in which chemical compounds responsible for malodor are transported and diffused from their emission sources to their potential receptors by means of atmospheric dispersion mechanisms. Aware that H₂S does not fully represent the odor impact problem by itself (in spite of it being probably the main responsible in the case of WWTP), the approach presented follows the same methodology as in classical air quality problems (and is somehow complementary to olfactometry, with some advantages and disadvantages), where pollution is evaluated through the analysis of individual species' behavior (NO₂, O₃, …).

In this context, knowledge about the source is a critical aspect for establishing emitter–receptor relationships (Hanna et al., 1981; Venkatram and Wyngaard, 1998). Acquiring this knowledge is generally complicated by the large number of pollution source points and the fugitive, unchanneled nature of pollution emissions, all of which make it difficult to determine emission rates a priori (especially for H₂S). Thus, an attractive approach to this problem is the use of indirect estimation methods for estimating the average emission from a wastewater treatment plant by comparing the ground-level concentrations experimentally measured using passive samplers with the modeling results for the same time period.

Overview

The experimental procedure presented herein uses the experimental measurements taken with passive samplers in the vicinity of wastewater treatment plants in combination with the numerical results of a diagnostic dispersion model to reveal some relevant aspects of the olfactory impact.

Moreover, diagnostic dispersion models also allow us to estimate the average emission rates of chemical species associated with routine wastewater treatment activities (in our case limited to hydrogen sulfide, one of the most relevant malodor-causing compounds in WWTPs).

For this study we had at our disposal a large number of measurements, both of certain meteorological parameters and of atmospheric concentrations of hydrogen sulfide, the chemical compound selected as the malodor tracer directly related to odor nuisance.

The study was carried out at five WWTPs in the Autonomous Community of Valencia (Spain), although for the exposition of the methodological details and the presentation of the results, we have preferred to focus on three of these plants, Conca del Carraixet (in Valencia), Rincón de León (in Alicante), and Grao de Castellón (in Castellón).

Available Measurements

Meteorological measurements

Meteorological towers were located at each of the WWTPs participating in this study. These infrastructures were each comprised of a 15-m-tall mast with wind measuring instruments (for speed and direction) at the top and temperature gauges at two heights (3 and 15 m, thus providing a direct measure of the surface thermal gradient); the system operates continuously and averages the data at 10-min intervals. Then, the average hourly wind speed, direction, and temperature values are used as inputs for the meteorological dispersion model, and the vertical temperature gradients and/or the sigma speed/direction values are used to estimate the atmospheric stability (U.S. Environmental Protection Agency [EPA], 2000).

Experimental ground-level concentrations measurements

The characterization of the hydrogen sulfide concentrations during the experimental campaigns was based on a passive sampler network distributed at different points around the plants. These measurement devices permit sufficient spatial coverage to be comparable with the results of the simulations, for which it is necessary to take into account the aerological characteristics of the site at the time of designing the experimental sampling net.

While there are many geometries, passive samplers usually consist of a small tube containing a diffusive substance (generally air) with one end closed and the other exposed to the environment (Campbell et al., 1994; Palmes, 1981; Proceedings of the International Conference Measuring Air Pollutants by Diffusive Sampling, 2001). By a process of molecular diffusion, the gas of interest is transported through the diffusive body from the open end to the opposite end. Here it is retained and eliminated by a fixing substance through a reaction specific to the gas of interest (De Saeger et al., 1995), thus establishing a concentration gradient that keeps the molecular pumping active. To estimate the average concentration of the pollutant in the atmosphere we presuppose a stationary molecular transport process, subject to Fick's second law,

(1)

where ms is the analyte mass incorporated (mol), D₁₂ the molecular diffusion coefficient of the analyte gas “1” (hydrogen sulfide) in another gas “2” (outside air), A the transversal section of passive sampler (cm²), c₁ the analyte concentration at the top of the passive sampler (mol/cm³), c₂ the analyte concentration at the bottom of the passive sampler (mol/cm³), t the sampling time (sec), and z the diffusion length, which ideally corresponds to the length of the passive sampler (cm). Because of the small amount of diffusion in this type of device, the sampling time needed is longer than in other measuring methods, varying from a few hours in highly polluted environments to several weeks in clean areas.

In this H₂S measurement work, passive samplers consisting of an acrylic plastic tube with a length of 7 cm and an inside diameter of 10 mm, closed at both ends by polypropylene stoppers, were used. Inside one of the stoppers were two stainless-steel meshes impregnated with a silver nitrate solution (Natush et al., 1975; Shooter et al., 1995; Tarasankar et al., 1986), which constitutes the absorbent medium.

The hydrogen sulfide becomes trapped in the impregnated meshes in the form of silver sulfur, according to the following reaction:

(2)

which is later analyzed by fluorimetry on the basis of the FMA (fluorescein mercuric acetate) “quenching” produced by the sulfur collected (Spedding and Cope, 1984).

As a general work norm for measuring in the field, three passive samplers were exposed on each site for five or six days, analyzing only two of them and designating the mean value as the concentration estimator; the third sampler was used only in the case of discrepancy between the other two. Lastly, all sampler concentrations were corrected for temperature, normalized at 20°C on the basis of the measurements from the meteorological station.

Dispersion model

Previous experiments with accumulated concentration measurements around different wastewater treatment plants showed a rapid decay in the accumulated concentrations with distance; thus, in the present study the simulation window was limited to the immediate vicinity of the plant, a square with 4-km sides, with the focus located approximately in its center.

The model used in the present work is based on the adjustment and integration of two independent codes: a diagnostic nondivergent meteorological model coupled to a puff diffusion model. The meteorological module incorporates the MATHEW base model (U.S. Department of Energy, Lawrence Livermore Laboratory) (Atmospheric Release Advisory Capability, 1997; ARAC, 1995). This is a regional, three-dimensional wind-field model that uses a variational technique to estimate the three components of the nondivergent wind field on a cubic grid, in consonance with the empirical starting measures. It considers complex topography explicitly (Espinós-Morató et al., 2008). It is site-independent, uses habitually available surface and aloft meteorological measures, and is computationally stable.

This module provides wind fields that advect the emissions by means of a transport and diffusion engine activated by an additional code, which is an adjustment of the EPA-INPUFF model (Petersen and Lavdas, 1986). The latter is a Gaussian puff diffusion model that allows instantaneous or continuous emissions from point sources to be simulated. It incorporates the Pasquill–Gifford formula to describe the dispersion and Briggs's algorithms to estimate the plume rise (Briggs and Haugen, 1975). The plume rise is calculated definitively for every “puff” at the moment of emission, with the vertical components of the wind field being responsible for the vertical movement during the rest of the dispersion process.

In selecting the codes, value was given to the concurrence between enough complexity for small-scale dispersive processes to be simulated and enough data-feeding simplicity for available local experimental information to be used.

The numerical executions were performed over variable time periods coinciding with the passive sampler exposure intervals and the hourly time resolution; the outputs consisted of ground-level concentration values on a grid that coincided with the topographic grid for each simulation time.

Experimental Methodology

Description of the Study Zone

The Castellón Grao WWTP (CAST code in Table 3, shown later) is located also in a flat crop area in Grao de Castellón that presents no prominent topographic restrictions, midway between the coast (to the east) and the town of Castellón (to the west) about 2 km away. This facility has a semi-urban profile with an average inflow of 42,768 m³/day (Figure 1A).

Figure 1. Location of the different WWTPs (coordinates are in Universal Transverse Mercator, UTM): (A) Grao de Castellón (CAST); (B) Conca del Carraixet (CR); (C) Rincón de León (RL).

The Conca de Carraixet WWTP (CR code in Table 3, shown later) is located in Alboraya, near the city of Valencia and close to the coast. It is located in the middle of a flat agricultural area, dominated by little farms. It is also a large-scale facility and treats sewage from a population of 109,520 inhabitants, with an average inflow of 38,962 m³/day. It has an urban/rural profile. The site is well ventilated and has no topographic obstacles (Figure 1B).

The Rincón de León WWTP (RL in Table 3, shown later) is located relatively close to the city of Alicante (Spain) and the coast. This large-scale facility, with an average inflow of 61,821 m³/day, has a clearly urban profile and services a population of approximately 667,000 people. It is situated near the sea (within 2 km of the coast in a straight line). Overall, the site is well ventilated and presents no major local topographic obstacles (Figure 1C).

The lack of important topographic obstacles at these three WWTP sites means that the local effects of the breezes will shape the distribution pattern of the ground-level concentrations.

Experimental Campaigns

The experimental measurements presented herein are the results of several campaigns using passive samplers distributed inside and outside the wastewater treatment facilities.

The distribution of the external measurement points was determined by the main wind direction frequencies in the study areas provided by the on site towers. Because most events in the three study areas occur during local breeze cycles, especially in the summer period, measuring points were placed at different distances according to their radial axes in the shape of arcs circumference (Figure 2).

Figure 2. Sampling grid of Grao de Castellón WWTP. On the right the scale corresponds to the topographic level (meters). The treatment plant location is in the middle of the window area.

Quality Control of Simulations

To reinforce the confidence in the results, the modeled values were compared with the experimentally available measurements, that is, the winds registered at the on site meteorological tower and the average concentrations provided by the passive samplers.

The use of a diagnostic model to obtain a three-dimensional meteorological field from local measurements has the advantage that the wind values obtained from the interpolated grid are perfectly similar to the meteorological measurements taken by the instrumentation at the same geometric point, assuring that advection of emissions are done by “real” winds.

To achieve best results, numerous executions of the wind field were carried out, adjusting some of the configuration model parameters.

Estimation of Average Emission Rates

The average rates were initially estimated using a unitary constant emission during the simulation period. Results were based on the comparison between the concentration values provided by the passive samplers and modeled values at the same geographic locations, integrated during the same exposure periods.

The comparison was carried out using the median of the population of ratios between the experimental measures and the model results. The result is a parameter representing the emission that best adjusts measurements to simulations and represents the average emission over the measurement time. The outputs presented herein are already affected by this factor and therefore can be assumed as experimentally adjusted simulations.

This factor (Q) is an indirect way to estimate the average emission rate of hydrogen sulfide from wastewater treatment plants.

The analysis of the differences between the measurements provided by the passive samplers and the results of the model was based on the calculation of various statistics to evaluate the degree and improvement of the adjust (see the Results section).

Proposal of an Emission Model That Includes Temporal Variability

The procedure just described provides constant emission estimation during the simulation period. Due to the nature of the emissions produced at a wastewater treatment plant, it seemed more realistic to try to introduce a temporal variation in the emission rate. This would improve the adjustment between simulations and experimental measurements and allow us to assess possible variations introduced in the environment under different emission conditions.

A general approach could be adjusted to the following model:

(3)

where is the time-dependent emission model proposed, the average emission level resulting from the adjustment between experimental and modeling results, the normalized function of temporal modulation emissions, which includes the contribution of environmental parameters, the normalized function of temporal modulation emissions, which includes the contribution of operational wastewater parameters, and the background emissions.

As an approximation to a temporal variability model, it seems reasonable to treat the generation of odor as a physical process of evaporation of the compounds of interest (in this case hydrogen sulfide) from the liquid medium to the atmosphere. Thus, the model described in this paper incorporates a temporal modulation of the environmental parameters (variations caused by plant operations were not considered at that stage). The scheme was:

(4)

The variable emission model proposed is based on Penman's formulation (Penman, 1948), later adapted by Quintanar et al., where the background emissions (R) were considered zero.

This formulation includes two independent contributions ( eq 5). The first has an aerodynamical basis, in which evaporation is due to turbulent transport of vapor by a process of eddy diffusion ( eq 6). The second approach is thermodynamical, where evaporation is the consequence of an increase in temperature due to incident radiation and the exchange of energy flows ( eq 7):

(5)

where

(6)

(7)

and where is the surface evaporation by the Penman model [kg m⁻² h⁻¹], is the saturation slope of vapor pressure curve for H₂S [k Pa K⁻¹], the psychrometric constant [k Pa K⁻¹], the net radiation [W m⁻²], the stored heat [W m⁻²], L the H₂S latent heat of vaporization [J kg⁻¹], the H₂S transfer coefficient for neutral conditions [nondimensional], the wind speed [m s⁻¹], the gas constant for dry air, 287.04 [J kg⁻¹ K⁻¹], the air temperature [K], the saturation vapor pressure of H₂S [k Pa], and the air pressure [k Pa].

Results and Discussion

Due to the extensiveness of the original experimental database, herein is shown only one complete study case, corresponding to an experimental campaign at the Grao de Castellón WWTP. For the evaluation of the results presented it has included the following information:

Table 1. Experimental and modeled concentration levels (for constant and variable emissions) at the Grao de Castellón WWTP

Node *	Experimentalmeasure	Constantmodel	Variablemodel
CE01	0.9 ± 0.2	1.9	0.4
CE02	0.3 ± 0.1	2.2	0.2
CE03	0.4 ± 0.0	2.0	0.4
CE04	0.6 ± 0.2	3.5	0.7
CE05	0.4 ± 0.1	5.0	0.4
CE06	0.8 ± 0.1	4.7	1.0
CE07	1.8 ± 0.1	1.7	1.8
CE08	1.3 ± 0.1	1.2	1.3
CE09	1.2 ± 0.1	1.8	0.8
CE10	0.6 ± 0.1	0.5	0.7
CE11	< LDM**	0.6	0.5
CE12	0.3 ± 0.0	0.8	0.4
CE13	0.2 ± 0.0	0.2	0.2
CE14	0.2 ± 0.0	0.2	0.2
CE15	< LDM**	0.4	0.2
Notes: *Nodes have the same name as in Figure 3. **LDM: Limit detection method. In our case is 0.2 ppb.

Figure 3. Temporal modulation during the simulation period for variable model at the Grao de Castellón WWTP (wind speed and direction values for the same period are also shown).

Figure 4. Least-squares adjustment for the two emission models used.

•	Temporal modulation introduced in the emission model for the simulation period (Figure 3).
•	Ground-level concentrations obtained from the experimental measurements and the model for the two emission scenarios considered (constant and variable) and their corresponding adjustment (Table 1 and Figure 4).
•	Various statisticals used to assess the most appropriate emission model (Table 2). Table 2. Calculus of BNMBF and ENMAEF for simulation with both emission models Statistic Constant model Variable model BNMBF 1.79 –0.07 ENMAEF 1.84 0.19
•	Adjusted average emission rates for the three WWTPs (Table 3).

Figure 4 shows the temporal modulation introduced in the emission model where the maximum emission introduced in the variable model coincides with the maximum hours of solar radiation. This fact favours daytime emissions (which are ultimately those that produce the most nuisance) and penalizes nocturnal emissions (where greater atmospheric stability forces higher concentrations in the vicinity of the WWTP).

Table 3. Estimated average emission rates for various campaigns in the three study zones

WWTP*	Campaign	Experimentalperiod (5 to 6 days ofexposure)	Averageemission(g/s)	Inflowwaterrate(m^3/day)
RL	I	February 2002	1.34	61,821
	II	July 2002	1.55
	III	January 2003	1.48
	IV	May 2003	2.29
	V	July 2003	2.47
	VI	January 2004	2.37
	VII	February 2004	1.03
CR	I	June 2002	1.74	38,962
	II	August 2002	0.44
	III	February 2003	0.09
	IV	April 2003	0.46
	V	June 2003	1.56
CAST	I	July 2005	1.36	42,768
	II	September 2005	1.41
Note: *See text for site codes.

Constant and time varying emissions have been normalized and are equal for the whole period. Table 1 shows the experimental and modeled ground-level concentrations for the campaign at Grao de Castellón and for each sampling point (Figure 2).

Adjustments between the experimental and model results (Figure 4) show the temporal modulation introduced by the variable model (around 0.8) adjusts better than one obtained by a constant model (around 0.2), with also a much higher correlation coefficient (R ² = 0.1779 and 0.826, respectively).

To assess which of the emission models represents more realistically the experimental ground-level concentration field, various statistical based on the performance of the dispersion models (Yu et al., 2006) were calculated. Finally, B_NMBF (normalized mean bias factor) and E_NMAEF (normalized mean absolute error factor), were selected (see Appendix for a detailed description).

Both metrics suppress the inflation in the results when it has low observational values while maintaining an adequate evaluation symmetry in the measurements. In both cases, zero value would be ideal.

Table 2 shows results of the two statistical parameters selected. According to the B_NMBF interpretation, in the present case study, the results indicate an underestimation in the observations by a factor of 0.9 using the variable model.

In the other hand, constant model overestimates observations by a factor around 3. In both cases the best values correspond to the proposed variable model. This result suggests that the introduction of a temporal dependence in emission improve the model behavior, and thus confidence in emission rates estimations. All the hydrogen sulfide emission rates presented in Table 3 are calculated according to the procedure just described. Only estimation from the variable model is shown.

According to the table of results, it can be seen that the average H₂S emissions for the three WWTPs range between 0.09 and 2.47 g/s. This fact implies differences of less than a factor of 4, which is lower, for example, than differences between treated water volumes at each plant. Despite the variability in the results shown, even for the same plant, this range is consistent between the plants studied. These averaged values suggest a total emission rate around 1 g/s for these facilities. This means that 50–100 kg of hydrogen sulfide is released into the atmosphere every day. This amount is about 8–10% of the incoming sulfur in a plant, potentially available to H₂S conversion, a result that agrees with standard values on urban wastewater quality.

The applicability of this methodology for estimating emission rates is limited in some cases by the environmental conditions present during the sampling period not providing a good relationship between experimental measurements and model values. This limitation can be overcome by increasing the spatial coverage of the measuring sampling grid so that no directions are left uncovered, or by increasing the exposition time of the passive samplers so that a greater variety of atmospheric conditions are included in the exposition periods.

Summary

This study has described a methodology for calculating H₂S emission rates, using an effective approach based on experimental measurements taken with passive samplers and numerical atmospheric dispersion simulations.

This approach has the advantage of evaluating plant emissions as a whole, including unchannelled emissions and emissions from diffuse sources, which is extremely difficult when dealing with individual emission sources.

The proposed procedure offers many advantages over other methodologies (e.g., olfactometric methods), providing an experimental base, simple to apply and at reasonable cost.

The estimation of the emission range in WWTPs studied was provided using this methodology. Availability of emission rates can be very useful for implementing corrective measures. Despite the variability of the results shown, even in the same plant, this range is consistent between the studied plants. These values suggest an average emission rate of around 1 g/s for these facilities. This means that a wastewater treatment plant releases about 100 kg of hydrogen sulfide per day into the atmosphere.

The procedure can still be improved in different ways; it was shown how estimations could be improved using a time varying emission model, with local and temporal parameters as inputs.

Lastly, this method has presented a way to temporally modulate the emissions from a plant by introducing the contribution of the various environmental parameters involved in the proposed emission model (Penman).

Appendix

The metric B_NMBF indicates both the magnitude of the factor between the modeled and observed quantities and the sense of the factor (positive value means that the model overestimates the experimental measures and negative means that the model underestimates them). The metric E_NMAEF means that the absolute gross error is n times the mean observation and model prediction.

Herein described are the analytical expression used for the metrics, where are the values of the model and are the values of the experimental measures, and the corresponding mean values (Table 4).

Table 4. Mathematical expression of the statistics used in this paper

Metric	Analytical expression
BNMBF (normalizedmean bias factor)	For For
ENMAEF (normalizedmean absolute error factor)	For For

Acknowledgments

This work is the result of the collaboration agreement between EPSAR (Entidad Pública de Saneamiento de la Comunidad Valenciana) and the Valencian Instituto Universitario CEAM‐UMH, having been financially supported by EPSAR. CEAM-UMH is partly supported by the Generalitat Valenciana and the projects GRACCIE (CSD2007‐00067, CONSOLIDER–INGENIO 2010 Program–Spanish Ministry of Science and Innovation) and FEEDBACKS (Prometeo/2009/006–Generalitat Valenciana).