Developing State-Wide Modeling Guidance for the Use of AERMOD – A Workgroup’s Experience Paper #1302 Carolee Laffoon, Joey Rinaudo, Raghu Soule, and Ted Bowie ENVIRON International Corporation, 8280 YMCA Plaza Dr., Bldg. 9, Baton Rouge, LA 70810 Chris Meyers Trinity Consultants, 4000 S. Sherwood Forest Blvd., Suite 503, Baton Rouge, LA 70816 Richard L. Madura RTP Environmental Associates, Inc., 285 W. Esplanade Ave., Suite 308, Kenner, LA 70065 Sirisak Patrick Pakunpanya Louisiana Department of Environmental Quality, Office of Environmental Assessment, Air Quality Assessment, P.O. Box 4314, Baton Rouge, LA 70821 ABSTRACT For over two decades, the Industrial Source Complex (ISC) dispersion model has been the primary model used to predict ambient air impacts from stationary sources, particularly in regard to regulatory considerations when permitting new or modified sources. Recent advances in dispersion modeling theory and computational power have led to major improvements over the ISC model. These improvements resulted in the creation of a new type of modeling algorithm referred to as the American Meteorological Society/U.S. Environmental Protection Agency Regulatory Modeling System (AERMOD). The U.S. Environmental Protection Agency (EPA) has proposed replacing ISC with the AERMOD dispersion model for evaluating near-field impacts for regulatory purposes, anticipating adopting it as the standard model by the end of 2004 and phasing out ISC in 2005. This action will have numerous implications for industrial facilities and regulators. AERMOD requires several additional geophysical meteorological input parameters that ISC does not utilize. Much discussion has taken place as to how these parameters should be set, but little overall guidance exists to assist modelers in determining their values. To address this lack of guidance, Louisiana Department of Environmental Quality (LDEQ) formed a Modeling Workgroup to evaluate issues with implementing AERMOD and to develop modeling guidelines for use with AERMOD within the state of Louisiana. Workgroup members performed several hypothetical case studies to (1) evaluate AERMOD’s behavior in comparison with ISC, (2) compare use of on-site meteorological data to National 1 Weather Service (NWS) data, and (3) determine AERMOD’s sensitivity to changes in input land-use parameters. The results of these analyses are being used to develop modeling guidelines, including site-specific parameters required by AERMOD. The paper summarizes the results of analyses performed by Workgroup members, as well as recommendations made by the members. In addition, general differences between ISC and AERMOD are discussed, including processing times, land-use parameters, meteorology inputs, and treatment of terrain. INTRODUCTION Regulatory History In the 1977 Clean Air Act (CAA), Congress mandated consistency in the application of air quality models for regulatory purposes, fulfilling the needs of industry and control agencies and encouraging the standardization of model applications. The Guideline on Air Quality Models (or simply Guideline), found in 40 CFR 51 Appendix W,1 was first published in April 1978 and was incorporated by reference in the regulations for the Prevention of Significant Deterioration (PSD) of Air Quality in June 1978. The Guideline was revised in 1986, updated with supplements in 1987, revised further in July 1993 in concurrence with being published as appendix W to 40 CFR Part 51, revised again in August 1995, and republished in August 1996. The Guideline is used by EPA, States, and industry to prepare and review air quality analyses requiring regulatory models such as new source review (NSR) permits and Sate Implementation Plan (SIP) revisions. The Industrial Source Complex (ISC) model, in various versions (the most current version is Industrial Source Complex Short Term Version 3 [ISCST3]), has served as the United States Environmental Protection Agency’s (U.S. EPA’s) basic regulatory model over the last two decades and has been extensively used to predict ambient air concentrations from industrial sources. Due to the various limitations and inadequacies of the ISC model recognized during the early years after its implementation, as described below, the American Meteorological Society (AMS) and the EPA initiated a formal collaboration in 1991 with the designated goal of introducing recent advances in handling boundary layer conditions.2 The AMS/EPA Regulatory Model Improvement Committee (AERMIC) formulated AERMOD and EPA proposed several changes to the Guideline, including (i) adopting AERMOD to replace ISCST3 as the regulatory model, (ii) revising ISCST3 by incorporating a new building downwash algorithm incorporating Plume Rise Model Enhancements (PRIME) and renaming the model ISC-PRIME, and (iii) updating the Emissions and Dispersion Modeling System (EDMS 3.1) in appendix A of the Guideline.3 In April 2003,4 EPA promulgated other previously proposed changes but deferred all of the aforementioned actions in order to address several significant public comments in response to the 2000 proposed changes. EPA later published a notice of additional information in response to public comments.5 Upon promulgation, it is anticipated that the EPA will allow a one-year transition period from the effective date, during which either of the two models (ISCST3 or AERMOD) can be used for regulatory-driven air quality modeling analyses. After the end of the transition period, AERMOD will be the recommended regulatory model to be used for regulatory air quality analyses. 2 Technical Description of Model Several shortcomings of the ISC model necessitated the need for developing a more sophisticated model.6 ISCST3 is best described as a Gaussian steady-state plume dispersion model with a minimum one-hour timestep, meaning the model calculates downwind concentrations for each hour of the modeled period treating the meteorological conditions as constant throughout the modeled domain and over time, ignoring the fact that the atmosphere is continuously varying spatially and over time and neglecting boundary layer effects. Other shortcomings of the ISCST3 model include poor characterization of building downwash and terrain features, a lack of chemical transformation and physical removal processes, and a useful range limit of approximately 40 to 50 kilometers. Simple terrain includes receptors with elevations below the top of the stack and at elevations above or below the stack base. Intermediate terrain includes receptors above stack top and below the plume centerline. Complex terrain includes receptors with elevations above the stack top. AERMOD was developed to overcome all of the aforementioned shortcomings. The AERMOD modeling system has 3 components: AERMOD - the air dispersion model; AERMET - the meteorological data preprocessor; and AERMAP - the terrain data preprocessor. EPA has described AERMOD as an advanced dispersion technique that incorporates state-of-the-art boundary layer parameterization techniques, convective dispersion, plume rise formulations, and complex terrain/plume interactions. Relative to ISCST3, AERMOD, as proposed, contained new or improved algorithms for: (i) dispersion in both the convective and stable boundary layers; (ii) plume rise and buoyancy; (iii) plume penetration into elevated inversions; (iv) treatment of elevated, near-surface, and surface level sources; (v) computation of vertical profiles of wind, turbulence, and temperature; and (vi) the treatment of receptors on all types of terrain (from surface up to and above the plume height). More recently, the PRIME algorithm was implemented into AERMOD to use the AERMOD meteorological profiles and to use the ambient turbulence intensities based on the AERMOD profiles.2 Reason for Workgroup Several states, including Louisiana, have set up local workgroups to: (i) conduct, compare, and discuss “case studies” to analyze the differences between ISCST3 and AERMOD including varying scenarios, i.e., meteorological data, land-use, source types, data point size (no. of sources), etc., (ii) assist the state in developing a strategy to implement the use of AERMOD in the state’s modeling guidelines, (iii) discuss issues, news, and/or questions on the AERMOD model, and iv) identify potential permitting and financial impacts on industry due to the differences between ISCST3 and AERMOD. Unlike for ISCST3, pre-processed “model-ready” meteorological data can not be used for AERMOD; data must first be processed with the AERMET pre-processor prior to use in AERMOD. In comparison with ISCST3, AERMOD requires several additional meteorological parameters, including albedo, Bowen ratio, and surface roughness, which are site-specific and need to be determined on a case-by-case basis based on a protocol recommended by the EPA. Since this can be confusing, the workgroup’s specific objectives were to review the use of these parameters, test the sensitivity of these parameters, and attempt to develop a state-specific protocol for determining the appropriate values to use for specific geographic/topographic locations. Among other topics that were reviewed by the workgroup was evaluation of the need to include digital elevation model (DEM) terrain data, since most of Louisiana is near sea level with flat terrain. 3 Activities Conducted by the Workgroup In order to address the goals of the Workgroup, the following analyses were performed by the workgroup: The first case-study compared concentrations predicted by ISCST3 with AERMOD-PRIME, for an urban setting with flat terrain and using National Weather Service (NWS) meteorological data, in order to compare concentration distribution, model run time, and isopleths. The sensitivity of the meteorological variables, i.e., albedo, Bowen ratio, and surface roughness was investigated. The model run-times were also compared. The second case-study accomplished the same objectives as the first case-study albeit comparing AERMOD-PRIME with ISC-PRIME for a rural setting with slight terrain. The model run times were also compared. The third case-study compared results from AERMOD-PRIME using NWS and on-site meteorological data. Also, a comparison of ISC-PRIME results from NWS and on-site meteorological data was performed. AERMOD-PRIME results were compared using no terrain and importing terrain at a relatively flat site. Issues related to the use of onsite meteorological data were identified. The model run times were also compared. The fourth case study performed similar comparisons to the first two case studies, albeit at a facility located in a marshy, rural setting and compared ISCST3 with AERMOD-PRIME. The model run times were also compared. Summary of Significant Findings The following observations were made by the workgroup in the course of conducting the aforementioned analyses: (i) AERMOD run-time is significantly greater than that of ISC models, (ii) more complex meteorological variables are used by AERMOD compared to ISC models making it difficult to predict the impact of variation in meteorological parameters on the concentrations, (iii) use of AERMOD meteorological parameters may not be appropriate for all averaging periods; it may be necessary to break runs up into seasons for shorter averaging periods, and (iv) AERMOD modeling generates significantly greater number of files compared to ISC models and is, in general, a significantly bulkier process than ISC. Two of the case studies indicated that AERMOD consistently produced lower maximum concentrations than ISC. The remaining two case studies, which used the same meteorological data set, indicated that AERMOD consistently produced higher maximum concentrations than ISC. The sensitivity analyses performed clearly indicated that surface roughness length is the most sensitive AERMOD meteorological parameter. Also, a “break-even point” was observed for this parameter, where the predicted concentrations were found to increase dramatically with the decrease in surface roughness length below a certain threshold. Specifically, modeled concentrations plotted against the surface roughness values indicated that there was a gradual increase in modeled concentrations with a decrease in surface roughness values until the value dropped below the 0.3 – 0.35 range, when the modeled concentrations increased by five to ten orders of magnitude. The second case study did not agree with some of the literature findings. 4 One of the literature studies9 concluded that the ratio of AERMOD to ISCST3 concentrations tends to increase as the length of the averaging period increases for sources in rural flat terrain. The second case study did not necessarily agree with this observation although the modeled terrain was moderately hilly (almost flat terrain). Another literature study10 concluded that for area sources, the concentrations predicted by AERMOD are approximately equal to those predicted by ISCST3 model. When the area sources were modeled as an independent group in the second case study, the differences between AERMOD and ISC were not insignificant. In the third case study, AERMOD predicted concentrations greater than ISC when utilizing on-site meteorological data for the annual averaging period and for the 24-hr averaging period with just one outlying meteorological year. The fourth case study concluded that the correlation between run time and number of receptors was not linear and also when seasonal periods are analyzed, meteorological pre-processing can be a time-intensive task. The results are covered in greater detail in the Results and Discussion section of this paper. LITERATURE REVIEW Several studies and observations presented in literature were reviewed by the Workgroup for the purpose of identifying previous similar analyses that have been performed and comparing to Workgroup observations. EPA conducted several studies to compare the AERMOD versus ISC analysis performed prior to the April 2000 Federal Register publication with the AERMODPRIME versus ISC analysis conducted after the Federal Register publication. In AERMOD: Latest Features and Evaluation Results,2 comparisons of model estimates with measured air quality concentrations for a variety of source types and locations were provided. An evaluation of AERMOD (version 02222) versus ISCST3 was performed for sulfur dioxide (SO2) for ten non-downwash databases in two phases: (i) the developmental evaluation, which was performed concurrently with model development for five databases, and (ii) independent evaluation to avoid any bias effects for five additional databases. Three short-term tracer studies and two conventional long-term monitoring databases in the developmental evaluation phase and one tracer study and four long-term monitoring databases in the independent evaluation phase were employed, in a variety of settings, for the purpose of generating observed concentrations. The databases included a variety of pollutant source types including, (i) flat terrain, moderately hilly terrain, hilly terrain, and grassy field, (ii) rural and urban environments, and (iii) near-surface non-buoyant and elevated buoyant releases. For the tracer databases, results for 1-hour averages were reported with the exception of one database, where 10-minute measurements were used. For the long term SO2 data sets, 3-hour, 24-hour, and annual results were reported. The EPA reran the AERMOD and ISCST3 models for all the aforementioned databases and generated the “ratio of modeled-to-observed Robust Highest Concentrations (RHCs).” The RHC served as a robust test statistic for assessing the difference between AERMOD and ISCST3 and represented a smoothed estimate of the highest concentrations, based on a tail exponential fit to the upper end of the concentration distribution. This procedure was used to reduce the effect of extreme values on model comparison. It was concluded from this study that (i) the performance of the revised version of AERMOD (02222) is slightly better than the April 2000 proposal and both versions of AERMOD significantly outperform ISCST3 as compared to monitored observations, and (ii) AERMOD (02222) with PRIME performs slightly better than ISC-PRIME for aerodynamic downwash cases. 5 Another document compiled by the EPA, Comparison of Regulatory Design Concentrations,7 documents a consequence analysis of effects on design concentrations and provides comparisons of design concentrations (on which emission control limits might be based) for a wide variety of source configurations and settings. There were three parts to this study: (i) the flat and simple terrain component; (ii) the building downwash component, and (iii) the complex terrain component. The flat and simple terrain consequence analysis was based on comparative runs made using a composite of standard data sets. These data sets include a range of point sources with varying stack parameters, area and volume sources, and two point sources in simple terrain. All source scenarios were evaluated with two meteorological data sets representing different climatic regimes in the U.S. For building downwash, a series of point sources with varying stack heights and different building configurations were included in the data sets. For the complex terrain, the study included a number of stack heights, buoyancy regimes, distances from source to hill, and hill types along with its own meteorological database. Observations from the “flat and simple terrain” analysis included: (i) AERMOD predicted lower than ISCST3 for low level stack in rural environments; (ii) AERMOD predicted higher than ISCST3 for taller stacks in rural environments for long-term averaging periods; (iii) AERMOD predicted lower than ISCST3 for urban short stacks and area sources for the short-term averaging period. In summary, the analysis indicated that: (i) for non-downwash settings, the revised version of AERMOD (02222), on average, tends to predict concentrations closer to ISCST3 with somewhat smaller variations than the April 2000 proposal of AERMOD, (ii) where downwash is a significant factor in the air dispersion analysis, the revised version of AERMOD predicts maximum concentrations that are very similar to ISC-PRIME; (iii) for those source scenarios where maximum 1-hour cavity concentrations are calculated, the average AERMOD predicted cavity concentration tends to be about the same as the average ISC-PRIME cavity concentrations; and (iv) in general, the consequences of using the revised AERMOD, instead of the older model ISCST3, in complex terrain remained essentially unchanged, although they varied in individual circumstances. Literature observations related to the Sensitivity of AERMOD to land-use parameters8 concluded that: (i) for surface sources, only surface roughness length affects the modeled concentrations significantly. The albedo and Bowen ratio have little or no effect on the annual modeled concentration. (The modeled short-term results for this study occurred at night when the albedo and Bowen ratio have no effect), (ii) for elevated stacks (case evaluated: 35 meters), all three land-use parameters affect modeled concentrations with albedo still having a relatively small effect and surface roughness still having the largest effect, and (iii) for very tall stacks (case evaluated: 100 meters), varying the land-use parameters had varied effect on the modeled concentrations. The authors of this study concluded that the effects these parameters have on the modeled design concentrations are sufficiently complex that it cannot be accurately anticipated what effect any change in those values will have on design concentrations for a given source configuration. The authors advise that reasonably accurate estimates of albedo, Bowen ratio, and surface roughness lengths are necessary for AERMOD to provide accurate results. A study6 related to the “run-time” issues and comparison of results to ISCST3 observed that the enhanced algorithms of AERMOD require significantly more computer time to run. AERMAP can, by itself, require a substantial amount of time to process and set up the modeling domain. AERMOD was observed to typically yield lower concentrations than ISCST3 when nearby complex terrain is present, but can yield higher concentrations in other terrain regimes. 6 Another published literature9 observed that AERMOD tends to predict lower concentrations than ISCST3 for shorter stacks (less than 20 meters) in rural conditions and that the ratio of AERMOD to ISCST3 concentrations tends to increase as the length of the averaging period increases for sources in rural flat terrain. Yet another study10 concluded that for point sources, the results varied depending on stack height, urban/rural, and terrain configurations; and for area sources, AERMOD’s predictions were almost identical to that of ISCST3. EPA and the American Meteorological Society (AMS) performed an evaluation of AERMOD with different met data sets in a 1998 study11 that showed the use of onsite data appeared to overpredict results, often by a factor of two. MATERIALS AND METHODS All model runs were performed with ISCST and AERMOD. Building downwash was obtained from either the Building Profile Input Program (BPIP) or the Building Profile Input Program with the PRIME downwash algorithm (BPIP-PRIME). Table 1 identifies the model specifications and hardware used in each test case. Table 1. Model and Hardware Specifications. Test Case Case Study 1 ISC 02035 with BPIP 95086 Software AERMOD 02222 with BPIPPRIME 95086 Case Study 2 01228 with BPIPPRIME 95086 03273 with BPIPPRIME 95086 Case Study 3 02035 with BPIPPRIME 95086 02222 with BPIPPRIME 95086 Case Study 4 00101 with BPIP 95086 02222 with BPIPPRIME 95086 Computers Used ISC - Mix 500, 900, and 1800 Mhz AERMOD - 1.8 GHz Dual Processor ISC – Dell Inspiron, 650 MHz CPU, 256Mb RAM AERMOD – Compaq Evo W6000 Workstation, Dual Xenon 2.0 GHz CPUs, 512 Mb RAM ISC – HP Vectra XE310, 1.2 GHz Celeron CPU, 512 Mb RAM AERMOD – Dell Dimension 8300, 3.2 GHz Pentium 4 CPU, 2 Gb RAM Dell Inspiron 8200, 2 GHz Pentium 4 CPU, 256k RAM Input meteorological and terrain data were obtained from a number of sources. The types and sources of data are shown in Table 2. 7 Table 2. Meteorological and Terrain Data. Test Case Case Study 1 Meteorological Data Type Source NWS Surface Air – Baton Rouge,LA Data Upper Air – Lake Charles, LA Case Study 2 NWS Data Surface Air – Shreveport, LA Upper Air – Longview, TX Case Study 3 NWS Data On-Site Data Surface Air – New Orleans, LA Upper Air – Slidell, LA River Region Environmental Association Hahnville, LA Station NWS Data Surface Air – New Orleans, LA Upper Air – Slidell, LA Case Study 4 Terrain Data Type ISC - none AERMOD none ISC AERMOD – 7.5 x 7.5 min DEM (1:24,000 scale) ISC - none AERMOD – 7.5 x 7.5 min DEM (1:24,000 scale) ISC - none AERMOD – 7.5 x 7.5 min DEM (1:24,000 scale) Source NA NA NA Geo Community Web Site11 NA Atlas Web Site12 Site, Source and Input Information Case Study 1 A hypothetical facility located in south Louisiana was modeled to determine the differences between AERMOD and ISC modeling results. This facility included information from various Louisiana companies and focused on PSD sources. Additionally, a sensitivity analysis was performed to determine the effect of varying AERMOD model inputs. The hypothetical facility center was located between Lafayette and Baton Rouge, Louisiana. The facility includes 47 point sources that are representative of PSD sources including boilers and flares. Stack heights for on-site sources range from 25 and 245 feet. 2916 receptors were included in the modeling domain. Meteorological data from 1997–2001 was used in both the AERMOD and ISC models. Surface air data from the Automated Surface Observation Station (ASOS) located at the Baton Rouge airport was used. Upper air data from the monitoring station in Lake Charles, Louisiana was used. Missing data was filled with meteorological data from the Baker, Louisiana monitoring site and the Capitol monitoring site located in Baton Rouge. 8 A Land Use/Land Classification (LULC) analysis was performed and the meteorological data was appropriately processed for use in the AERMOD model. The ISC modeling was performed using urban land-use classification parameters. Additionally, a series of AERMOD model runs were set up by varying albedo, Bowen ratio, and surface roughness values from their respective minimum and maximum values to determine what effect, if any, these parameters would have on AERMOD model results. The BPIP-PRIME algorithm was used in the AERMOD model analysis. The standard BPIP algorithm was used in the ISC model analysis. A hypothetical pollutant was modeled at 3-hour, 24-hour, and annual averaging periods. Maximum short-term emission rates were used for all sources in the short-term analyses (and what about long-term averaging perid? - the annual average emission rate..). Case Study 2 A hypothetical facility located in Northern Louisiana was modeled to determine the differences between AERMOD and ISC modeling results. The facility is located in a marshy valley between a river and a bayou. Elevation changes of up to 60 feet are present in the surrounding area. Additional model runs were setup to determine what effect, if any, varying the AERMOD meteorological parameters would have, as well as the effect of running the AERMOD model with and without terrain data. The hypothetical facility includes five point sources, three area sources, and fifty-two volume sources. Point sources included in this analysis are representative of two large boilers, two cyclones, and a cogeneration facility. Stack heights for on-site sources range from 50 to 66 feet. Area sources represent two log piles and a bark hog. Fugitive emissions from unpaved roads are included as a series of volume sources. A regional emissions inventory including 20 point sources representing miscellaneous combustion sources, cyclones, and fugitive emissions was also included for both the AERMOD and ISC modeling analyses. 934 receptors were included in the modeling domain (it is also important to state the interval/distance of separation between rceptors).. Meteorological data from 1990-1994 was used in both the AERMOD and ISC models. Surface air data from the National Weather Service Station located at the Shreveport airport was used. These data were collected prior to the installation of the ASOS. Upper air data from the monitoring station in Longview, Texas was used. Surface air data was complete, but upper air data was completed according to U.S. EPA guidelines using interpolation. An LULC analysis was performed and the meteorological data was appropriately processed for use in the AERMOD model. The ISC model was performed using rural land-use classification parameters. U.S. Geological Survey (USGS) maps were used to perform the land-use analyses. Additionally, a series of AERMOD model runs were setup by varying albedo, Bowen ratio, and surface roughness from their respective minimum and maximum values to determine what effect, if any, these parameters would have on AERMOD model results. USGS 7.5 min x 7.5 min DEM Data (1:24,000 scale) was used in both the AERMOD and ISC models. AERMAP was used to import terrain data for all model objects in the AERMOD- 9 PRIME model and to generate hill-height scale data that was used to drive advanced terrain processing algorithms. Additionally, a second set of AERMOD models was set up without importing the terrain data with AERMAP to determine whether or not the use of this data would affect model results. The BPIP-PRIME algorithm was used in both AERMOD and ISC modeling analyses. A hypothetical pollutant was modeled at 3-hour, 24-hour, and annual averaging periods. Maximum short-term emission rates were used for all sources in the analyses. Case Study 3 A hypothetical facility was located in Southeastern Louisiana along the bank of the Mississippi River near a meteorological data monitoring station that served as the source of the onsite meteorological data used in some of the model runs for this case study. The main goals of this effort were to determine the differences in utilization of AERMOD vs. ISC with varying meteorological data and dispersion coefficients. NWS meteorological data and site-specific meteorological data were used as the data sources in the meteorological data comparison. The dispersion coefficient scenario compared rural dispersion coefficients and urban dispersion coefficients in ISC to represent the “heat island” effect occasionally assumed for Louisiana facilities. A third goal of this scenario was to determine the effects of running an AERMOD model with the inclusion of terrain data for an area that is relatively flat, as opposed to excluding the terrain data. The hypothetical facility contained a total of forty-four point sources and three area sources. These sources are representative of combustion devices, particulate control devises, storage tanks, wastewater treatment equipment, and fugitive emissions. Hypothetical structure downwash effects were also included with the scenario, as this is a common situation with most facilities of this type. The BPIP-PRIME algorithm was used to estimate these effects in both the ISC and AERMOD model runs. The receptor grid for this scenario was laid out according to current LDEQ modeling guidelines of 100 m receptor spacing along the facility fence line and a Cartesian grid with 100m receptor spacing out 1km from the facility and 1km receptor spacing out 10km from the facility. For the AERMOD model runs, terrain data was included such that the entire modeled area accounted for its appropriate elevation. The terrain data for these analyses was obtained from USGS DEMs. It should be noted that the terrain in Southeastern Louisiana is relatively flat; in fact the most significant topographical feature within the receptor grid of these analyses is the levee system that runs along the Mississippi River. An LULC analysis of the area was performed utilizing the most recent aerial photographs available along with a general knowledge of the topography of the area. This LULC analysis provided a basis for estimation of the seasonal albedo, Bowen ratio, and surface roughness coefficients. This data was then appropriately processed along with the different sets of meteorological data for inclusion in the AERMOD models. Meteorological data from 1998-2002 was used in both the AERMOD and ISC models. Surface air data from the ASOS located at the New Orleans International Airport and upper air data from 10 the Slidell Weather Service Meteorological Observatory (WSMO) were used in the NWS scenarios. The onsite met data used was collected from the River Regional Environmental Association station located in Hahnville, Louisiana. Ground level concentrations of a hypothetical pollutant were then estimated for the 3-hour, 24hour, and annual averaging periods. Maximum short-term emission rates were used for all sources in the analyses, as appropriate. Case Study 4 A hypothetical manufacturing facility with two nearby facilities located in open marshland south of New Orleans was modeled to determine the differences between AERMOD and ISC modeling results. The domain overlapped Universal Transverse Mercator (UTM) Zones 15 and 16. The site is characterized as open, treeless marshland with relatively no elevation change and alternating land and water. This location was chosen in order to evaluate the availability and quality of data for a remote region, and to determine what effect, if any, including terrain data would have on the AERMOD model results. The hypothetical facility included ten point sources and two area sources. Sources typical in this area of the state would include boilers, gas turbines, tanks, and other miscellaneous combustion sources. Stack heights for on-site sources range from 4 to 70 feet. Area sources represent fugitive emission sources. A regional emissions inventory including two point sources at an adjacent facility to the southwest and six point sources at a facility located approximately two kilometers to the southwest were also included for both the AERMOD and ISC modeling analyses. 2.25 million receptors were included in the modeling domain, simulating a large PSD modeling project. Meteorological data from 1998-2002 was used in both the AERMOD and ISC models. Surface air data from the ASOS located at the New Orleans International Airport was used. Upper air data from the monitoring station in Slidell, Louisiana was used. Where necessary, all data was completed according to U.S. EPA guidelines using interpolation. An LULC analysis was performed and the meteorological data was appropriately processed for use in the AERMOD model. The ISC model was performed using rural land-use classification parameters. USGS maps were used to perform the LULC/Land-use analyses. Albedo, Bowen ratio, and surface roughness were calculated seasonally based on values provided in the AERMOD Manual for water and swamp. USGS 7.5 min x 7.5 min DEM Data (1:24,000 scale) was obtained from the Atlas web site12 and used in the AERMOD model. As the area is essentially flat, no terrain data was used in the ISC model. AERMAP was used to import terrain data for all model objects in the AERMOD-PRIME model and to generate hill-height scale data that was used to derive advanced terrain processing algorithms. The BPIP-PRIME algorithm was used in the AERMOD modeling analysis, and BPIP was used in the ISC modeling analysis. A hypothetical pollutant was modeled at 3-hour, 24-hour, and annual averaging periods. Maximum short-term emission rates were used for all sources in the analyses. 11 RESULTS AND DISCUSSION The sites modeled by the Workgroup were intended to provide a representative sampling of the various terrain types located in Louisiana, which include marsh, swamp, coastal plains, coastal savannah, river valleys, and rolling terrain. Prior to the performance of AERMOD, it was necessary to run the AERMET preprocessor to obtain input files for the meteorological data. In order to obtain the surface parameters, an LULC analysis must be performed using available information relevant to the terrain, land use, population density, and other physical characteristics. The LULC analysis can be performed by a number of means, including use of an area-weighted average, a distance-weighted average, or other appropriate methodology. All case studies performed by the Workgroup used an area-weighted approach to the LULC analysis. Based on the LULC analyses performed for the individual sites, site-specific surface parameters were developed for each site and used in the base case AERMOD runs for comparison to ISC. In addition, several studies varied the surface parameters, generally from the minimum to maximum values presented in the tables provided in the AERMOD user’s manual. Table 3 shows the annual average surface parameters calculated for the four test cases and the ranges of surface parameters used in the studies. Table 3. Surface Parameters used in AERMET. Test Case Case Study 1 Case Study 2 Case Study 3 Case Study 4 Albedo Bowen Ratio Site: 0.21 Site: 2.24 Min/Max: 0.05-0.95 Min/Max: 0.1-10 Site: 0.2 Site: 0.62 Min/Max: 0.05-0.95 Min/Max: 0.1-10 Site: 0.15 Site: 0.65 Min/Max: 0.13-0.17 Min/Max: 0.2-1.26 Site: 0.16 Site: 0.175 Surface Roughness Site: 0.74 Min/Max: 0.001-1.3 Site: 0.34 Min/Max: 0.001-1.3 Site: 0.44 Min/Max: 0.14-0.77 Site: 0.0813 In general, USGS topographic maps were used to develop the LULC. One observation made is that USGS topographic maps for remote locations in Louisiana are not updated as frequently as those containing developed areas, and some of the topographic maps referenced for coastal Louisiana were out of date. Due to the significant coastal erosion occurring in Louisiana’s coastal wetlands, it is likely that major differences exist between the information presented in the topographic maps and the terrain currently present. Aerial photos of coastal Louisiana are available from several sources, including LaCoast13 in conjuction with the USGS National Wetlands Research Center. However, these aerial photos are updated infrequently and may not be representative of conditions currently existing. In these cases, the most recent data available was used for the LULC analyses. DEM files for the AERMAP preprocessor were readily available from a number of sources, and several sources were used, including the GeoCommunity web site11 and the Atlas web site12. DEMs from the Atlas site are provided in the “new” format, and the CRLF program in the AERMAP package (demfilz) was necessary to convert the files into the proper input format. 12 Some DEMs for the remote site south of New Orleans were not available from the Atlas site, and AERMAP was performed without a DEM for this location. In addition to the efforts required to prepare and perform the AERMET and AERMAP preprocessors, a point of interest of the Workgroup was the computer resources necessary to run AERMOD as compared to ISC. AERMOD was found to take four to thirteen times longer to run than ISC, with run times ranging from 1.5 hours for small receptor grids to 22.9 hours for large receptor grids. Considering the additional time required to prepare and perform the preprocessors, overall preparation and modeling time can be expected to significantly increase the time required to prepare and run AERMOD. Table 4 shows the typical run times for ISC and AERMOD under the various test cases. Table 4. Comparison of Run Times. Test Case Case Study 1 Case Study 2 Case Study 3 Case Study 4 *ISC-PRIME 3-hr ISC AERMOD 21 min 4.1 hr 22.1 min* 2.1 hr 20-30 min* 1.5 hr 3.5 hr 22.1 hr Averaging Period 24-hr ISC AERMOD 21 min 4.1 hr 22.1 min* 2.1 hr 20-30 min* 1.5 hr 2.5 hr 22.9 hr Annual ISC AERMOD 18 min 4.5 hr 22.1 min* 2.1 hr 20-30 min* 1.5 hr 1.67 hr 22.5 hr All four cases performed model runs with both ISC and AERMOD using NWS meteorological data, with Case Study 3 also utilizing local meteorlogical data. Table 5 shows maximum off-site concentrations in g/m3 obtained in the model runs using the calculated annual average surface parameters and NWS meteorological data. The test cases fell into two categories-sites where ISC predicted greater concentrations than AERMOD and sites where AERMOD predicted greater concentrations than ISC. Case Studies 1 and 2 both demonstrated lower maximum concentrations from AERMOD than from ISC, with the exception of a single meteorological year in Case Study 2. Case Studies 3 and 4 both demonstrated higher maximum concentrations from AERMOD than from ISC, with the exception of one averaging period in a single meteorological year in Case Study 4. Two possible causes have been identified explaining this difference. First, Case Studies 3 and 4 both used New Orleans surface meteorological data and Slidell upper air data. Both stations are relatively near the coastline, which appears to influence the modeling results. Second, Case Study 4 incorporates an extremely low surface roughness factor due to the prevalence of open water and marsh in the area. However, Case Study 3 incorporates a medium surface roughness factor, possibly indicating that meteorology in the coastal area plays a greater role than surface factors in AERMOD. Interestingly, the isopleths in Case Study 4 were generally smaller for the AERMOD runs than for the ISC runs for all averaging periods, consistent with Case Studies 1 and 2, in spite of AERMOD predicting higher maximum off-site concentrations. 13 Table 5. Maximum Receptor Concentrations Using NWS Meteorological Data. Test Case Case Study 1 Case Study 2 Case Study 3 Case Study 4 Year 1997 1998 1999 2000 2001 1990 1991 1992 1993 1994 1998 1999 2000 2001 2002 1998 1999 2000 2001 2002 3-hr ISC 243.2 229.4 256.9 220.73 225.2 664.7 520.2 476.3 485.3 410.6 128.33 114.10 142.54 139.40 125.06 1700.51 1921.68 5679.37 4486.71 3155.76 Averaging Period 24-hr Annual AERMOD ISC AERMOD ISC AERMOD 134.9 92.8 53.2 17.2 12.0 133.5 92.0 37.8 18.2 12.2 117.9 86.9 39.6 17.6 11.7 91.63 79.3 38.4 17.3 11.6 107.4 85.8 38.1 20.3 13.4 714.1 173.4 205.6 25.1 30.5 521.5 133.1 116.3 29.4 23.9 348 153.8 131.4 28.5 23.9 512.0 160.6 99.1 21.5 17.0 394.4 145.9 114.8 29.0 23.5 159.38 665.00 492.37 15.65 35.45 205.15 556.34 782.27 16.39 36.10 184.83 640.54 635.19 17.32 33.04 237.95 624.18 778.95 18.89 59.29 169.88 562.88 622.60 14.92 33.65 2494.82 408.50 641.64 18.68 25.43 3588.53 404.83 647.70 21.60 29.29 3274.34 750.91 514.28 19.17 26.25 644.79 20.59 31.71 406.14 20.52 - Case Study 3 also performed ISC and AERMOD runs using on-site meteorological data obtained from the River Regions Environmental Association meteorological station located in Hahnville, LA, approximately 10 miles west-southwest of the New Orleans International Airport station. Table 6 shows the difference between the ISC results when on-site meteorological data is used as compared to NWS data for a single receptor. The on-site data produced concentrations ranging from approximately 78-150 % greater than the NWS data. Table 7 shows the difference between the AERMOD results when on-site meteorological data is used as compared to NWS data for a single receptor. As with the ISC runs, the on-site data produced significantly greater concentrations than the NWS data, ranging from approximately 245-480% greater. An earlier EPA/AMS study11 showed that the use of on-site meteorological data overpredicted results by as much as 200%. In this case, the station used was capable of collecting data in a single layer, possibly leading to the higher concentrations. 14 Table 6. Comparison of NWS vs. On-Site Meteorological Data in ISC. Year 3-hr 1998 1999 2000 2001 2002 Average Difference NWS 347.95 340.16 304.45 300.43 318.52 On-Site 665.00 556.34 640.54 624.18 562.88 89.91% Averaging Period 24-hr NWS On-Site 74.58 128.33 79.69 114.10 84.88 142.54 65.70 139.40 64.33 125.06 Annual NWS On-Site 7.24 15.65 6.64 16.39 6.77 17.32 6.00 18.89 7.28 14.92 77.95% 147.78% Table 7. Comparison of NWS vs. On-Site Meteorological Data in AERMOD. Year 3-hr 1998 1999 2000 2001 2002 Average Difference NWS 196.12 167.36 187.09 214.76 198.63 On-Site 492.37 782.27 635.19 778.95 622.60 246.83% Averaging Period 24-hr NWS On-Site 67.57 159.38 53.49 205.15 48.48 184.83 41.21 237.95 59.68 169.88 Annual NWS On-Site 6.83 35.45 6.70 36.10 6.74 33.04 6.45 59.29 7.58 33.65 272.53% 482.23% The Workgroup was tasked with evaluating impacts resulting from variation of the surface parameters input to AERMET. Case Studies 1 and 2 performed these evaluations. Table 8 shows the results of the sensitivity analyses performed on the surface parameters in Case Study 1, and Table 9 shows the results of the sensitivity analyses performed on the surface parameters in Case Study 2. In both cases, albedo and Bowen ratio were found to have little to no impact on the maximum off-site concentrations. However, surface roughness was found to significantly affect the results, with concentrations increasing from two to eight times for the 3-hr and 24-hr averaging periods when comparing low surface roughness to high surface roughness. Interestingly, while concentrations for the annual average period in Case Study 2 were consistent with the other averaging periods, concentrations were lower for low surface roughness in Case Study 1. It is possible that this anomaly can be attributed it to possible meteorological or sitespecific terrain effects. It is also possible that the high stacks in this case study may have impacted this result. 15 Table 8. Surface Parameter Sensitivity Analysis for Case Study 1. Averaging Period 3-hr 24-hr Annual AERMOD Input Parameter Albedo Bowen Ratio Surface Roughness 0.05 0.95 0.1 10 0.001 1.3 134.9 134.9 134.9 134.9 413.4 133.2 52.9 54.0 53.4 53.1 101.1 50.4 12.3 8.4 9.9 12.2 8.9 15.2 Table 9. Surface Parameter Sensitivity Analysis for Case Study 2. Averaging Period 3-hr 24-hr Annual Albedo 0.05 0.95 714.1 714.1 204.4 231.1 30.1 36.4 AERMOD Input Parameter Bowen Ratio Surface Roughness 0.1 10 0.001 1.3 714.1 714.1 1729.0 213.4 207.4 205.8 507.5 105.5 30.6 30.0 56.5 19.5 Louisiana is comparatively flat, and it is common to not incorporate terrain data in ISC runs due to the gradual, if any, variations in elevation. However, the Workgroup was interested in evaluating the performance of AERMOD using the non-default FLAT terrain option as compared to incorporating DEM data. Case Study 3 performed this evaluation, and Table 10 shows the effects of running AERMOD with terrain and with the non-default FLAT terrain option. The FLAT terrain option was found to produce concentrations ranging from 20.9840.90% greater than runs incorporating DEM data. Case Study 2 also evaluated the effects of running AERMOD with terrain and the non-default FLAT terrain. It was concluded from Case Study 2 that the higher concentrations in the flat terrain case were found to be due to a small difference in distance from receptor to centerline. There was a source-receptor elevation difference of 1.35 meters for the max receptor in the model using the DEM data; whereas, the elevation difference was only 1 meter in the flat terrain model since the terrain and receptors were at equal heights. Table 10. Comparison of FLAT vs. DEM Terrain Data. Year 1998 1999 2000 2001 2002 Difference 3-hr Flat DEM 492.37 343.26 782.27 380.96 635.19 292.16 778.95 406.82 622.60 492.10 40.90% Averaging Period 24-hr Flat DEM 159.38 119.57 205.15 99.56 184.83 91.66 237.95 126.73 169.88 126.13 39.87% Annual Flat DEM 35.45 31.52 36.10 31.82 33.04 29.58 59.29 29.72 33.65 26.36 20.98% 16 CONCLUSIONS AND RECOMMENDATIONS Performance of AERMOD was found to be a bulkier and more time consuming process than the performance of ISC. A significantly greater number of files are generated by the preprocessors that must be properly managed in order to successfully complete a model run. Modeling time for the cases presented was significantly greater using AERMOD, and, if seasonal periods are analyzed, it is anticipated that preprocessing itself can become time and labor-intensive. Terrain and meteorological data were readily available. However, DEM data is difficult to obtain for remote areas of the Louisiana coastal area, and USGS topographic maps are potentially inaccurate when considering coastal erosion. A review of the most up-to-date aerial/satellite photos or other data may be necessary when performing the Land Use/Land Classification Analysis. In general, Case Studies 1 and 2 predicted concentrations from AERMOD lower than those obtained from ISC. Case Studies 3 and 4 predicted higher concentrations from AERMOD than those obtained from ISC. The major difference between the two groups was the meteorological data sets used. Case Studies 3 and 4 used New Orleans surface data and Slidell, LA upper air data. Compared to the stations used in the other two cases, these stations are near the coast, which may possibly influence the results. It is possible that these results are meteorological data set dependent. Likewise, on-site meteorological data used in Case Study 3 produced higher concentrations than National Weather Service Data when running AERMOD. The quality and representativeness of the on-site data should be evaluated prior to use in AERMOD runs. Only the surface parameter of surface roughness was found to significantly impact the results of an AERMOD run for the locations and meteorological data sets modeled. Albedo and Bowen ratio had little to no impact on results. Great care must be taken when performing the Land Use/Land Classification analysis to properly characterize the surrounding terrain and, in particular, surface roughness in order to avoid over- or under-predicting concentrations. As a result of this work, the Workgroup made the following recommendations for consideration by regulatory agencies: Develop a means of establishing agreement with regulatory agencies on meteorological variables is essential before any modeling can begin Establish guidance for on-site meteorological data, i.e., when is it appropriate to use, what minimum data is required, etc. Establish guidance on the characterization of areas for surface parameter land use analysis, possibly including acceptable ranges for various locations in the state and differentiation among rural, urban, and residential areas Establish guidance on use of surface parameters in wet/dry/average conditions and incorporation of seasonal variability 17 Establish guidance on the methodology of sector surface parameter calculation for development of surface parameters (i.e., area-weighted, distance-weighted, etc.) Establish guidance for terrain data inputs, including identification of acceptable data sources and accounting for missing terrain files. ACKNOWLEDGEMENTS Patrick Pakunpanya with LDEQ assembled and led the Workgroup. Ryan Clausen and Tom Petroski with URS performed the modeling analysis for Case Study 1. Raghu Soule, formerly with Trinity Consultants (currently with ENVIRON International Corporation), Chris Meyers with Trinity Consultants, and Lyn Tober formerly with Trinity Consultants (currently with Providence Environmental Engineering and Environmental Group, LLC) conducted the Case Study 2 modeling analysis. Ted Bowie, Doug Daugherty, Carolee Laffoon, and Joey Rinaudo with ENVIRON performed the modeling analysis for Case Study 3. Rick Madura with RTP conducted the Case Study 4 modeling analysis. Other Workgroup members included John Black with Enviro-One, Cade Borque with Shaw Group, Kerry Brouillette with URS, Kevin Calhoun with CRA, Scott Dorris with ERM, Beth Hughes with C-K and Associates, Keith Jordan and Tien Nguyen with LDEQ, Gerhard Pringer with Tulane, and Yousheng Zeng with Providence. REFERENCES 1. SCRAM Website. See http://www.epa.gov/scram001/guidance/guide/appw_03.pdf (accessed July 2004). 2. AERMOD: Latest Features and Evaluation Results, U.S. EPA, Office of Air Quality Planning and Standards, Emissions Monitoring and Analysis Division, EPA-454/R-03-003, June 2003. 3. 65 FR 21506, Requirements for preparation, Adoption, and Submittal of State Implementation Plans (Guideline on Air Quality Models), 40 CFR Part 51, April 21, 2000. 4. 68 FR 18440, Revision to the Guideline on Air Quality Models: Adoption of a Preferred Long Range Transport Model and Other Revisions, April 15, 2003. 5. 68 FR 52934, Availability of Additional Documents Relevant to Anticipated Revisions to Guideline on Air Quality Models Addressing a Preferred General Purpose (Flat and Complex Terrain) Dispersion Model and Other Revisions. September 8, 2003. 18 6. Reeves, Douglas, Understanding and Adapting to New Dispersion Models, Trinity Consultants. 7. Comparison of Regulatory Design Concentrations, AERMOD vs ISCST3, CTDMPLUS, ISCPRIME, U.S. EPA, Office of Air Quality Planning and Standards, Emissions Monitoring and Analysis Division, Research Triangle Park, NC 27711, EPA Report No. EPA-454/R-03-002, July 2003. 8. Grosch, T.G. and Lee, R.F., Sensitivity of the AERMOD Air Quality Model to Selection of Land Use Parameters. 9. PES Website, http://home.es.com/aerfaqs.htm (accessed 2004). 10. Peters, W.D., et al., Comparison of Regulatory Design Concentrations: AERMOD Versus ISCST3 and CTDMPlus, DRAFT, April 1999. 11. Cimorelli, Alan J., et al, Minimum Meteorological Data Requirements for AERMOD – Study and Recommendations, Draft Document, December 1998. 12. GeoCommunity Website. See http://data.geocomm.com/dem/demdownload.html (accessed 2004). 13. Atlas Website. See http://www. atlas.lsu.edu (accessed 2004). 14. LaCoast Website. See http://www.lacoast.gov (accessed 2004). KEYWORDS Dispersion modeling; AERMOD; ISC; PRIME; Louisiana 19