Comparison of AERMOD and SYMOS'97 models for calculating odor dispersion: A study case in Uttenweiler

on . . Hits: 13639

sesion04 souza00In this article we make a comparative analysis of the performance of the AERMOD and SYMOS’97 dispersion models for simulating the dispersion of odorous compounds in the atmosphere. Although there are a large amount of air dispersion models in literature, there are few data bases that allow the assessment of results obtained by the dispersion simulation of odorous compounds.

F. F. C. Souza, L. Hoinaski, H. de Melo Lisboa1

 

1Laboratory of Air Quality Control, Department of Sanitary and Environmental Engineering, Technological Center, Federal University of Santa Catarina, Campus Reitor João David Ferreira Lima, Trindade, Florianópolis, SC, Brasil, PO Box 476

 

 

   Competing interests: The author has declared that no competing interests exist.

   Academic editor: Carlos N Díaz.

   Content quality: This paper has been peer reviewed by at least two reviewers. See scientific committee here

   Citation: F. F. C. Souza, L. Hoinaski, H. de Melo Lisboa., Comparison of AERMOD and SYMOS'97 models for calculating odor dispersion: A study case in Uttenweiler, Ist International Seminar of Odours in the Environment, Santiago, Chile, www.olores.org

   Copyright: 2014 olores.org. Open Content Creative Commons license., It is allowed to download, reuse, reprint, modify, distribute, and/or copy articles in olores.org website, as long as the original authors and source are cited. No permission is required from the authors or the publishers.

   Keyword: AERMOD, SYMOS’97, gaussian dispersion models, odors.

Download presentation

Video

Abstract

   In this article we make a comparative analysis of the performance of the AERMOD and SYMOS’97 dispersion models for simulating the dispersion of odorous compounds in the atmosphere. Although there are a large amount of air dispersion models in literature, there are few data bases that allow the assessment of results obtained by the dispersion simulation of odorous compounds. The goal of this work is to assess how well the AERMOD model is able to predict odorant concentrations in the nearby’s of a stack. Thus, a comparative analysis of the performance of AERMOD and SYMOS’97 models was accomplished for the Uttenweiler experiment. In order to determine AERMOD’s best potential, in this particular case, we used the measurements of an ultrasonic anemometer and considered the building downwash effect of the obstacles located in the stack vicinity of the pig farm. We also used a peak-to-mean concentration approach to adjust the model to predict peak concentrations. The comparison of observed and modeled concentrations demonstrated the importance of assessing a simulation, allowing the detection of error patterns. The AERMOD model tends to under predict the concentrations for this experiment presenting a factor of two of 23.1%, while SYMOS’97 reaches 57.7% for the same index. The detection of the AERMOD’s under predicting pattern allowed the improvement of its results. Through this article we intend to contribute to the better understanding of the dispersion modeling of odorous compounds in the atmosphere.

Introduction

   The assessment of olfactory nuisance is an expensive and highly complex task. Among the options for accomplishing this task, the modeling of odorant dispersion becomes increasingly interesting and, in some cases imperative, as in assessing the impact of not yet established activities and in proposing policies for land use and occupancy.

   The dispersion modeling of odorous compounds presents some particularities and challenges. The odorant perception occurs in a few seconds, whereas most models simulate only average concentrations of greater time intervals (De Melo Lisboa et al. 2006). An alternative to this problem is the use of peak-to-mean ratios, which allow the conversion of average concentrations of greater time duration to mean concentrations of shorter duration. Moreover, the dispersion of odorous compounds may show synergistic and masking effects, which are complex to model.

   These and other uncertainties make it essential to carry out validating studies of the dispersion simulation of odorous compounds. The validation may allow the simple verification of the model’s predicted results or even allow the comparison of model performance which would identify the best suitable for simulating certain situations or compounds.

   In the present work we made use of the dispersion experiment of Uttenweiler, Germany (Bächlin; Rühling; Lohmeyer, 2002) in order to compare the performace of models of odorant compounds. This experiment was chosen because it has already been tested which allows the comparison with previous results, beyond presenting data of high quality and quantity. The experiment still accounts with the downwashing influence of obstacles due to a low stack, which makes the experiment representative, of the dispersion of odorous compounds (Rørdam et al. 2005)

   AERMOD is a gaussian dispersion model established by the USEPA as a regulatory model for the simulations of atmospheric pollutants of up to 50 km from the source. The model simulates simple and complex terrains and may include the influence of obstacles through its PRIME algorithm (EPA – Environmental Protection Agency 2004). This paper aims to conduct a comparative study between the AERMOD and SYMOS'97 dispersion models. SYMOS’97 is a gaussian dispersion model, established as a regulatory model for the dispersion of odors in Czech Republic. In order to intercompare the model performance of AERMOD and SYMOS’97, the methodological procedure proposed by Keder in adapting the model SYMOS'97 to odors will be repeated (Keder, Bubnik, and Macoun 2003; Keder 2008a), this time with the AERMOD model. Both simulations made use of the Uttenweiler experiment data set.

Materials and Methods

   The object of this study is a pig farm with a barn, located in Uttenweiler, state of Baden-Württemberg, Germany. The farm and its surroundings are constituted primarily of cultivated fields and a small forest of approximately 80 x 80 meters, located at the north of the barn. The barn consists of two buildings respectively of 7.65 and 10.65 meters high. The smaller one has two stacks of 8.5 meters of height connected to the internal ventilation systems. Only one of the stacks was used in the experiment. The one used presented three shafts totaling an area of 3.6 m² (Aubrun and Leitl 2004; Bächlin, Rühling, and Lohmeyer 2002).

   A set of 15 experiments, including a preliminary one, was performed. Each experiment lasted 10 minutes during which small volumes of SF6 tracer gas and odorant gas were released in parallel. In the leeward side of the stack SF6 sensors were positioned next to 11/12 trained assessors. A sonic thermo-anemometer made measurements of temperature, wind speed and direction every 10 seconds. Meanwhile, two tracer detectors quantified the SF6 concentration while the assessors wrote down the perceived intensity on a scale of 0-5, along the same time intervals. The experiment was conducted on December 12th and 13th of 2000 and on October 31st of 2001. All experiments had enough cloud coverage to prevent turbulent conditions, and the winds weren’t in the calms range (Bächlin, Rühling, and Lohmeyer 2002).

   Keder, Bubnik e Macoun (2003), and Keder (2008) accomplished the adaptation and validation of the model Symos'97 to odors with the odorant Uttenweiler experiment. SYMOS'97 is a gaussian dispersion model that makes use of five stability classes developed by Bubnik and Koldovsky for Czech Republic. In order to adapt the model to odors, the odorant and tracer experiments were simulated and a peak-to-mean ratio of 2.2 was applied (Freeman and Cudmore 2002) at the hourly concentrations calculated by the model for both pollutants. Since the Uttenweiler experiment did not provide an intensity to concentration odor correlation, we adopted the following procedure to find the best fit curve that correlates these variables. The estimated model concentrations were grouped by classes of intensity levels, according to the intensity level perceived by the assessors. The median concentrations of each intensity class were calculated for the experiments B, C and E to O. Finally, the logarithmic equation that best describes the set of points was obtained. In order to compare the results obtained by Keder, for SYMOS'97, with the results obtained with the AERMOD model, the same procedure described above was adopted.

   The AERMOD model is composed of three basic modules: the terrain processor, AERMAP; the micrometeorological processor, AERMET; and the dispersion module, AERMOD. The model assumes steady state conditions for each time simulated and is not suitable to simulate calm conditions. In order to meet the minimum model requirements the meteorological data of two ground stations were merged with the data of a sounding station. The thermo-anemometer’s 10 minute average wind and temperature were used for simulating the experiments. The remaining variables were obtained from the meteorological station of Laupheim, located 22 km northeast of the farm. The soundings were obtained from Schnarrenberg airport, located 84 km northwest of the site.

   A correction of the values of wind speed and direction was made in the third stage of AERMET, the model rounds the wind directions to every 10 degrees, and also rounds the wind speed values to classes of wind established in the model. The values of standard deviation of horizontal wind - SAnn (degrees) and the standard deviation of the w component of the wind - SWnn (m/s) were also inserted in the third stage. These deviations were calculated from the sonic anemometer’s data set measurements for each experiment.

   A digital elevation map SRTM3 with a 90 meters resolution was used for the extraction of altitudes. This option was chosen due to the presence of some inconsistencies in the experiment’s registered altitudes. The building downwash caused by three obstacles were included in the simulation: two barn buildings and a small forest area of 20 meters of altitude. We adopted an albedo of 0.18, a Bowen ratio of 0.4 and a roughness length of 0.01.The chosen values of albedo and Bowen ratio were taken from the model’s meteorological manual for conditions of wet cultivated soil in autumn in the northern hemisphere. For the roughness length the same value used in the GRAL – Graz Lagrangian Model – validation with the Uttenweiler experiment was chosen (Pongratz, Öttl, and Uhrner 2012).

   The intercomparison of results from both models was divided into two stages: tracer experiment and odorant experiment. We used the statistical indices proposed by Chang and Hanna (2005) in the evaluation of atmospheric dispersion models, i.e., bias (Bias), fractional bias (FB), normalized mean square error (NMSE), correlation coefficient (R), geometric mean bias (GM), geometric variance (VG) and a factor of two (FAC2), as well as the acceptable limits for these indices in model assessment:

sesion04 souza01

where Co is the observed value, Cp is the model prediction, C̄ is the given average of the data set and σp is the standard deviation of the data set.

   Once carried out the intercomparison of the tracer experiment results, two fit curves of intensity concentration were tested: the curve obtained by Keder with the SYMOS'97 model for the Uttenweiler experiment and the curve obtained by repeating the procedure of Keder with the AERMOD model. In order to evaluate which of the curves shows better performance in converting AERMOD’s concentrations to intensities the same statistical indices described above was used. Lastly, a comparative plot between the two previous curves, along with a third theoretical curve that also correlates intensity and concentration for swine buildings was done.

 

Results

   Reapeating Keder’s procedure for the odor validation of SYMOS’97, it was found that the AERMOD model tended to underestimate the simulated concentrations. The SYMOS'97 model acceptably reproduced the measured concentrations of the Uttenweiler experiment for the factor of two, while the AERMOD model was not able to reproduce the same performance (Fig 1). An analogous result was obtained by Vieira De Melo et al. (2012), in which the AERMOD tended to underestimate the concentrations near the simulated source in the wind tunnel direction of 220 degrees, on a small-scale reproduction of the Uttenweiler experiment.

sesion04 souza02 sesion04_souza03.png

Fig 1 : Comparison of the simulated and observed concentrations for the factor of two with the AERMOD (left) and SYMOS’97(right) models. The image on the right was adapted from Keder (2008b).

   In terms of statistical indices, SYMOS’97 showed better performance than the AERMOD (Table 1). Since AERMOD’s estimated concentrations underestimates the measured concentrations for the field experiment, it was decided to test the application of another peak-to-mean ratio that maximizes the concentration values estimated by the model. The tested ratio of value 6.6 showed improvement in the comparison of the simulated and measured concentrations for the indices of bias, fractional bias, normalized mean square error, correlation, geometric mean bias, geometric variance and factor of two.

   After accomplishing the tracer experiment performance comparison, the intensity–concentration curve obtained in this experiment was compared with Keder’s fit curve. It was found that the curve obtained in this work was able to adjust properly the concentrations to intensities obtaining better agreement than on the tracer experiment (Table 2).This is due to the fact the odorant dataset is wider than tracer dataset, which helps diluting errors. The curve obtained by Keder also showed good fit to the set of odorant concentrations calculated by AERMOD since it exhibits similar behavior to the curve obtained in this work. It should be noted that both curves mask the errors arising from the dispersion model itself. An evidence of this is the fact that the curve obtained in this work provides better fit to the data simulated by AERMOD than the curve obtained by Keder. Comparing the curves obtained in this work and Keder’s curve with a theoretical one (NICOLAI et al., 2000 apud YU, 2010) it was found that the first two have a faster growth, while the theoretical curve presents a softer growth, ie, higher odorant concentrations would be required to elevate the odorant intensities in the theoretical curve (Fig 2).


Table 1: AERMOD and SYMOS’97 comparsion for the Uttenweiler tracer experiment. Source: Adapted by the author from KEDER (2008b).
 

AERMOD

PM 2.2

AERMOD

PM 6.6

SYMOS’97

PM 2.2

Perfect Model

(Chang and Hanna 2005)

Limits

(Chang and Hanna 2005)

BIAS

19.34

-1.63

6.08

0.0

<0.3

FB

0.96

0.05

0.18

0.0

<0.3

NMSE

26.58

0.77

0.60

0.0

<1.5

R

0.67

0.67

0.74

1.0

Close to 1

MG

4.35

1.45

0.88

1.0

0.7<MG<1.3

VG

53.66

7.11

1.02

1.0

<4

FAC2 (%)

23.1

53.8

57.7

100.0

>50

The fit equations of intensity-concentration obtained in this experiment, in Keder’s experiment and a theoretical curve are presented below:

Iod = 2.378(Cod)0.3223 - this article;

Iod = 1.086(Cod)0.464 - Keder, Bubnik and Macoun 2003);

Iod = 1.57(log10Cod) - 0.466 - (Nicolai et al., 2000);

where Iod is the odorant intensity and the Cod is odorant concentration. The theoretical curve was not used in the calculation of statistical indices because it presents negative values for concentration values between 0 and 2.

sesion04_souza04.png

Fig 2: Comparison of the intensity - concentration curves.

 

Table 2: Performance comparison between Keder’s curve and this articles curve for the intensity fitting of AERMOD’s calculated concentrations.
 

This article

KEDER

Perfect Model

(CHANG; HANNA, 2005)

Limits

(CHANG; HANNA, 2005)

BIAS

0.73

1.34

0.0

<0.3

FB

0.30

0.63

0.0

<0.3

NMSE

0.32

0.75

0.0

<1.5

R

0.59

0.55

1.0

Close to 1

MG

1.91

0.74

1.0

0.7<MG<1.3

VG

2.79

22.48

1.0

<4

FAC2 (%)

66.7

45.3

100.0

>50

Conclusions

   The importance of accomplishing validating studies for the dispersion of odorant pollutants has been clearly stated in this work. The comparison of observed and modeled concentrations allowed the detection of an under-prediction pattern of AERMOD for the Uttenweiler experiment, and a correction procedure of this pattern was tested. SYMOS’97 showed superior predictive capability of concentrations near the stack for this experiment. The use of a higher peak-to-mean ratio was successfully used to correct under predicting patterns of modeled concentrations. The use of wind deviations concentrated the plume dispersion along its centerline reducing its lateral dispersion. The AERMOD’s odorant simulation showed better fit than the tracer experiment due to the dilution of modeling errors in a wider dataset and to the use of the fit curve.

References

Articles in magazines:

Aubrun, S.; Leitl, B. 2004. Unsteady characteristics of the dispersion process in the vicinity of a pig barn. Wind tunnel experiments and comparison with field data. Atmospheric Environment, v. 38, n. 1, p. 81–93.

De Melo Lisboa, H. et al. 2006. Dispersion of odorous gases in the atmosphere - Part I: Modeling approaches to the phenomenon. The Science of the total environment, v. 361, n. 1-3, p. 220–8.

Keder, J. 2008a. Adaptation fo the Czech regulatory dispersion model for odour dispersion calculation, its validation and critical evaluation. Chemical Engineering Transactions, v. 15, p. 17–22.

Keder, J.; Bubnik, J.; Macoun, J. 2003. Validation of czech reference model symos’97, adapted for odour dispersion modeling 1. p. 139–142.

Vieira de melo, A. M. et al. 2012. Modelling of odour dispersion around a pig farm building complex using AERMOD and CALPUFF. Comparison with wind tunnel results. Building and Environment, v. 56, p. 8–20.

Books:

Bächlin, W.; Rühling, A.; Lohmeyer, A. 2002. Bereitstellung von validierungs-daten für geruchsausbreitungs-modelle-naturmessungen. (Baden-Württemberg, Germany).

Chang, J. C.; Hanna, S. R. 2005. Technical Descriptions and User’s Guide for the BOOT Statistical Model Evaluation Software Package. (USA).

EPA – Environmental Protection Agency. 2004. User’s guide for the AMS/EPA regulatory MODEL-AERMOD. n. September, p. 216.

Freeman, T.; Cudmore, R. 2002. Review of Odour Management in New Zealand. (New Zealand).

Pongratz, T.; Öttl, D.; Uhrner, U. 2012. Documentation of the Lagrangian Particle Model GRAL (Graz Lagrangian Model). (Germany).

Rørdam, H. et al. 2005. Regulatory odour model development: Survey of modelling tools and datasets with focus on building effects NERI Technical Report No. 541.(Denmark).

Yu, Z. 2010. Development of a Livestock Odour Dispersion Model. University of Saskatchewan. (Canada).

Web pages:

Keder, J. 2008b. Czech odour dispersion model and its validation. Available at: <http://www.cschi.cz/odour/files/pdf1/003-Keder-CHMI.pdf>.

Comments (0)

There are no comments posted here yet

Leave your comments

  1. Posting comment as a guest.
Attachments (0 / 3)
Share Your Location

Please note that this site uses cookies in order to work properly.

See more about our cookie policy Learn more

I understand

Please read the following to learn more about our cookies policy:

 

What are cookies?

   A cookie is a text file stored in a user’s web browser on any device they use to access a website that holds information regarding the user’s visit, such as preferences. When the user returns, the browser provides the cookie with the stored information to the site.

What cookies are used for?

   Cookies are used for adjusting a website’s content to fit a user’s preferences and optimize the website. They store useful information that improve the user’s experience of a website. They are most commonly used for:

  •     Recognizing the type of device a user is browsing with and any preferences applied to optimize the website specifically for the device.
  •     Creating statistics that help website owners to understand how their users interact with their website, which allows them to improve their structure and content.

What types of cookies are used?

   There are two types of cookies: persistent cookies and session cookies. Persistent cookies remain on your hard drive for a period of time specified in the cookie’s file parameters or until removed manually. When you return to a website and it requires you to login again despite previously storing your login information, it is usually because the persistent cookie expired; this helps to increase security while maintaining accessibility.

   Session cookies, on the other hand, are used temporarily and expire once the website or browser is closed. They are used to track user activity on a website during a single visit. When a website requires that you verify your age or location once every visit before allowing you to view content and without requiring additional personal details, that is a session cookie at work.

Do cookies include personal data?

   If there is a need for the collection of personal information, such as for creating accounts, then cookies may store personal information. However, it is required by data protection law that users are informed of the collection of personal data. This data will also be encrypted to render it inaccessible for unauthorized users.

Managing cookies

   By default, browsers are configured to accept cookies. However, these settings may be changed to block cookies entirely, or to inform the user each time they are used. Detailed information about cookies and the options associated with them are available in each browsers’ settings.

Which cookies does collect olores.org?

   Olores.org collect cookies for 2 purposes:

  • Register statistical data.
  • Set language preferences.

   In addition we use third party cookies through Statcounter to collect different data.

StatCounter Analytics Cookies

   StatCounter is a web analytics service. We use StatCounter to track activity on our website. These stats help us to understand how people are interacting with our website and to improve the design and functionality of our site so that we can offer a better online experience to our visitors. If you visit olores.org, a StatCounter analytics cookie (called "is_unique") may  be placed in your browser.  This cookie is used only to determine whether you are a first-time or returning visitor and to estimate unique visits to the site. No personal information is stored in the cookie.

Refuse Statcounter cookies.

You may set your browser to refuse/accept StatCounter analytics cookies by clicking here.

NOTES:

    • Your decision to refuse/accept StatCounter analytics cookies applies to all websites which use the StatCounter service (including the StatCounter site itself).
    • If you refuse all StatCounter analytics cookie, a refusal cookie (called "refusal_cookie") will be set to remember this preference and any existing StatCounter analytics cookies in your browser will be destroyed.
    • If you delete/remove/destroy the refusal cookie, you must revisit this page in order to re-set your preference.
    • The refusal cookie is set only for your current browser and machine. If you use multiple browsers/machines, you must set a refusal cookie in each case.
    • You can also change your cookie settings directly in your browser. Learn more about cookies and how to manage them here: http://www.allaboutcookies.org/cookies/index.html
    • Or you can learn about how to adjust cookie settings for specific browsers here: