Dispersion Modeling of SO 2 Emissions from a Lignite Fired Thermal Power Plant using CALPUFF

In this work, dispersion of sulfur dioxide (SO2) in the vicinity of Mae Moh power plant, the largest fossil fuel power plant in northern Thailand, was investigated using well known air dispersion model. The area of 2,500 km around the plant was studied, with spatial resolution of 200 x 200 m. Publicly available MM5 and CALMET software were used to provide meteorological conditions within the study domain, while CALPUFF was used to simulate the patterns of SO2 dispersion, based on actual plant operations in winter, summer and rainy seasons of the year 2009. Comparison against measurements from monitoring stations was made. Simulated results were found to agree qualitatively and quantitatively well with measured data. Root mean squared errors were found in the range between 2.19 to 8.32 μg/m. The CALPUFF model can be used for SO2 dispersion prediction with satisfactory accuracy.


Introduction
Over 75% of Thailand's total annual lignite production comes from Mae Moh open-pit lignite mine, located in northern Thailand.All of them are consumed by Mae Moh power plant, producing electricity to serve about 10% of the total electrical energy demand in the country.Ten production units of Mae Moh power plant comprise four units of 150 MW and six units of 300 MW.The total generating capacity is 2,400 MW, consuming lignite at approximately 17 million tons per year.Mae Moh lignite is generally of low quality, with high sulfur content (3%).Consequently, the power plant generates significant amount of sulfur dioxide (SO 2 ).High concentrations of SO 2 have negative impact on fauna and flora, and can produce acute injury in the form of foliar necrosis, even after relatively short duration exposure and.To reduce SO 2 emissions, flue gas desulfurization (FGD) plants have been installed to all power units since 2000 and their removal efficiencies are around 97-98% (Jonathan & Lovell, 1998).Average SO 2 emission rate after FGD was about 450 g/s.Monitoring of air pollutants is performed continuously in the area surrounding Mae Moh power plant with total of 11 stations (Jitman, 2010).From operation and management points of view, it would be good idea to predict and forecast SO 2 dispersion around the power plant when there are changes in operation of the plant, geometry of the area or local atmospheric conditions, so that measures can be implemented to prevent any harms that may arise in the future.
Since SO 2 can travel for a long distance according to wind direction and meteorological conditions, a detailed long-range transport model is needed to simulate the atmospheric processes and the resulting pollutant concentrations in the air.California puff model (CALPUFF) is one of the most popular dispersion models to study long-range pollutant transport and is recommended by the US Environmental Protection Agency (EPA) for regulatory applications (Scire et al., 2000).It is a non steady-state, Lagrangian dispersion model that provides significant improvement in atmospheric motion simulation, especially across large horizontal distances (i.e., 50 km or more) when compared to the traditional steady-state, Gaussian dispersion models.
There have been a number of works adopting CALPUFF to simulate pollutant transport and dispersion.Varna and Gimson (2002) modeled the inter suburb dispersion of particulate matter pollution in Christchurch, New Zealand using CALPUFF during a wintertime particulate pollution episode.Levy et al. (2002) applied the CALPUFF atmospheric dispersion model with meteorological data from the National Oceanic and Atmospheric Administration's rapid update cycle model to a set of nine power plants in Illinois to evaluate primary and secondary particulate matter impacts across a grid in the Midwest, USA.Cohen et al. (2005) developed a regression model to approximate the CALPUFF predicted concentrations and to determine the impacts of roadway proximity on ambient concentrations of three hazardous air pollutants.Jiang et al. (2003) employed a back trajectory method using the CALPUFF model in a reverse diffusion mode to investigate the details of ozone formation within the Puget Sound area of western Washington state, USA.Bandyopadhyay (2009) studied ambient air quality around an industrial estate in India and predicted the ground level concentration of SO 2 under various scenarios using industrial source complex short term (ISCST3) model.
In Thailand, Doolgindachbaporn (1995) used CALPUFF with a fumigation algorithm grid model to evaluate maximum hourly, maximum daily, and average ground level SO 2 concentrations to predict the impact from Mae Moh power plant.The simulated concentrations were found to slightly overestimate the measured data.This was due to assumption of no chemical transformations and depositions of SO 2 .Kanokkanjana (2004) modeled dispersion of air pollutants (SO 2 , NOx, and PM10) in Laem Chabang industrial estate using ISCST3 and CALPUFF models.Good correlation was obtained between 50 predicted maximum one-hour average concentrations and observed data.However, poor correlation for all hourly data was obtained.Correlation between the ISC and CALPUFF model outputs was good between the 50 maximum concentrations.According to the time series analysis, the model outputs did not agreed well with the monitoring data because the quality of emission inventory was not appropriate for modeling.Jiruagnimitsakul (2004) used CALPUFF air quality model to sulfur dioxide emission from a coal power plant where majority of predicted concentrations agreed well with the observed values.Tuaycharoen (2006) compared the model performances between regional atmospheric modeling system (RAMS)-California meteorological model (CALMET)/CALPUFF and ISCST3.They were used to simulate the dispersion of pollutants from Khanom power plant, the biggest thermal power plant in southern Thailand.Two different seasons were simulated; during December 2004 under the influence of north-easterly monsoon, and during June 2005 under the influence of south-westerly monsoon, respectively.Simulated NOx concentrations from both models were compared.The RAMS-CALMET/CALPUFF was shown to perform better than the ISCST3 model.Patima (2006) investigated SO 2 dispersion in Laem Chabang industrial complex using CALPUFF.Gas dispersion in this area was influenced by prevailing wind and atmospheric conditions In this work, CALPUFF was utilized to predict SO 2 dispersion around Mae Moh power plant.Seasonal variation was simulated for January, March and October 2009, representing winter, summer and rainy seasons.Comparison between the model simulation and the measurement results from monitoring stations was performed to evaluate prediction accuracy for high and low concentration episodes of SO 2 .

Study Domain
In this study, dispersion of SO 2 emitted from the power plant over Mae Moh basin was evaluated.The entire study domain (50 km x 50 km) is shown in Figure 1, with the power plant positioned at the center.The southwest corner of the domain is located at longitude 99.5E, latitude 18.1N.The northeast corner is located at longitude 100.0E, latitude 18.5N.There are 11 receptor points in both the X and Y directions, with a square grid spacing of 200 m.All 11 monitoring stations are indicated in Figure 1, surrounding the power plant.Their names and locations are given in Table 1.
Table 1.Locations of 11 monitoring stations of the Mae Moh power plant Figure 1.Terrain and locations of monitoring stations in the modeling domain

Source Characteristic
The emission source for this study is from Mae Moh power plant, the largest fossil fuel power plant in northern Thailand.This plant has 10 boilers with total capacity of 2400 MW.There are eight stacks at the plant.Details are shown in Table 2.

Emission Monitoring Stations
Each monitoring station contains multiple gaseous pollutant analyzers for SO 2 , NO 2 , PM10, TSP, CO, and O 3 measurement.Concentration data from these monitoring stations are available online, and can be accessed in real time to the power plant supervisory team.Under the circumstances where SO 2 concentration has tendency to exceed the alarming level, the team might consider shutting down some power generators as a countermeasure to prevent excessive atmospheric level of SO 2 .
Meteorological measurements are also routinely undertaken at the monitoring stations.These measurements include wind speed, wind direction (Figure 2), ambient temperature, surface pressure and relative humidity (Table 3).In this study, these meteorological data were used as input to CALMET model, and SO 2 concentrations were used for comparison with the model simulation results.Simulation was performed for 1-31 January, 1-31March and 1-31 October 2009, totaling 63 days.

Meteorological Modeling
From the power plant previous experience in 2009, high concentrations of SO 2 were observed to occur in January, March and October for winter, summer and rainy seasons, respectively.The concentrations were the highest, among data recorded during the past five years.These months were therefore chosen as representatives for this simulation study.The days used were those with the highest emissions monitored.
CALMET meteorological model was employed for this study because it is designed to produce hourly fields of three-dimensional winds and various micrometeorological variables based on the input of routinely available surface and upper air meteorological observations.The model requires geophysical data including gridded fields of terrain elevations and land use categories (defining surface roughness, albedo, Bowen ratio, etc.).In this case, the gridded geophysical data, including land use category and terrain height were generated from USGS terrain and land use database.Meteorological data for January, March and October 2009 from the 11 monitoring stations and data set from the meso-scale meteorological model (MM5) were used as input to the model to generate the diagnostic meteorological field.The 11 vertical layers were set at heights of 20,40,80,140,240,420,700,1100,1700, 2500 and 3600 m.

SO 2 Dispersion Modeling
CALPUFF was used in this study to simulate long-range transport of SO 2 in the atmosphere.The CALPUFF modeling system includes three main components: CALMET, CALPUFF, and post processing and graphical display programs (EPA, 1998).In this work, the modeling was performed to simulate SO 2 dispersion in three season episodes.The modeling results were then compared with the measurement data.

Statistical Data Analysis
Application of various statistical parameters was undertaken and discussed by Elbir (2003) in comparison between predicted and observed data sets.Similar evaluation was carried out in this work to evaluate the performance of our model.The following parameters were calculated; root mean squared error (RMSE) and its components (systematic and unsystematic RMSE), correlation coefficient, and index of agreement.They are defined as: where N is the number of data, P i is the model prediction value, O i is the observed concentration, and O is the mean value of the observed data.σ O and σ P are standard deviations of the observed and predicted data.i P ˆ is derived from the least square regression of P i and O i , respectively.

Meteorological Patterns
Typical wind vectors over the study area in a representative winter day (7 January 2009) were presented in Figure 3.It can be seen that the wind came from the north-eastern direction.Analysis of all grids in the domain by CALMET modeling revealed that in the early morning at 06:00, wind speed at 110 m level was in the range between 0.01 -14.06 m/s, and mixing height was 50 -1097 m, while at Mae Moh power plant, mixing height was 50 m, just under stack exit.At noon, wind speed at 110 m was 0.01 -3.59 m/s, and mixing height was 50 -1759 m, meanwhile at the power plant, mixing height was about 300 m.Because of low wind speeds and relatively small mixing heights over stack, accumulation of SO 2 could potentially occur as air dispersion was rather limited by stable atmospheric condition.Meteorological conditions in the evening (18:00 and 23:00) were similar to those in the early morning at 06:00.

SO 2 Spatial Distribution
Dispersion of SO 2 was mainly dependent upon wind direction and strength.It was found that in the early morning (06:00) and late evening (23:00), there were large air movements from north-eastern direction.The mixing height was only 50 m, just below the stack outlet.Under this circumstance, accumulation of SO 2 was unlikely to occur.Therefore, the concentrations were very low at both times.At midday, the wind speed blowing over the domain was found to be very low and the mixing height was slightly higher than the plant stack height.Under these conditions, capping of gaseous plume could occur, preventing the SO 2 dispersion to the upper atmosphere.Figure 4 shows the predicted spatial distribution of SO 2 at 12:00 on a representative winter day.High concentrations of SO 2 were found in the downwind areas close to the power plants.Figure 5 shows the comparison between the model results and monitoring data from all stations at the same time of day.The predicted SO 2 concentrations were found to agree well with the measurement, with standard deviation of 8.16 µg/m 3 .

SO 2 Temporal Variation
Figures 6 -8 show daily variation of SO 2 concentrations at locations of three monitoring stations; namely Ban Sop Moh, Pratupha Army Camp, and Mae Moh Government Center throughout the months of January, March and October, respectively.High concentration episodes (≥ 10 µg/m 3 ) were evident.The predicted daily average values of the SO 2 concentrations were also found to follow similar temporal patterns to measurement values.For absolute value, simulation results appeared to slightly over-predict when compare to monitored data.Stable atmospheric stratification, calm (low wind speed), and presence of a ground-base inversion were normally encountered during winter in Mae Moh basin.These were unfavorable meteorological conditions for efficient mixing of pollutants, giving rise to possible high concentration episode.
For the case of low concentration episodes (< 10 µg/m 3 ), the predicted and measured daily averages of the SO 2 concentration at the three monitoring stations were compared (shown in Figures 9-11).The model appeared to show large variation in this case, especially during monsoon with rain and precipitation.

Statistical Analysis
Table 4 presents the mean, maximum and standard deviation, together with other statistical parameters for the measured and predicted daily SO 2 concentrations in the year 2009.The index of agreement (d) was found to vary from 0.0 (theoretical minimum) to 1.0 (perfect agreement between the observed and predicted values).The value of d should be close to 1.0 and total RMSE should be close to 0.0 for good prediction.From this work, statistical comparison between the model results and the measured data from six monitoring stations showed moderate degrees of agreement ranging from 0.36 to 0.73.Standard errors in terms of total RMSE were found in the range between 2.19 to 8.32 µg/m 3 .Analysis of RMSE indicated comparatively small systematic error, with unsystematic error approaching the RMSE.Overall, quantitative evaluation of the model results revealed that the simulated model performance was very good, especially for winter and summer seasons.
Table 4. Statistical analysis of the predicted and observed time series of SO 2 concentration

Conclusion
This paper presents simulation of sulfur dioxide dispersion from Mae Moh power plant, and comparison between the predicted and measured concentrations in the Mae Moh basin.Both measured and predicted data included daily SO 2 concentrations from three seasons of the year 2009.For all monitoring stations, the predicted SO 2 concentrations were found to agree qualitatively well with the measured data.Quantitatively, several overpredictions were observed, especially during rainy season.These differences might be caused by the influence of complex area from mountain terrains in Mae Moh basin, and uncertainties of measured data and meteorological data used in the model.Our findings showed that the CALPUFF model can be used for the SO 2 dispersion prediction with approximate accuracy.The model appeared to be a valuable tool in evaluating air pollutants dispersion from Mae Moh power plant to its vicinity.

Figure 2 .
Figure 2. WindRose of January, March and October 2009

Figure 3 .
Figure 3.Typical wind vector on a representative winter day

Figure 4 .Figure 5 .
Figure 4. Predicted spatial distribution of SO 2 at noon on a representative winter day

Table 2 .
Stack characteristics of the Mae Moh power plant