Air Dispersion Modeling Expert Witness Services (AERMOD, ALOHA, AERSCREEN, CALPUFF, & AIHA Box Models)

RoseyFields LLC — Paul E. Rosenfeld, Ph.D., Principal Forensic Environmental Chemist

Technical reference page for attorneys, judges, consultants, regulators, and scientists.

1. What Dr. Rosenfeld Does

Dr. Paul E. Rosenfeld reconstructs the concentration, timing, and spatial extent of human exposure to airborne contaminants — usually for periods in the past, often decades in the past, and almost always at locations where no one was monitoring at the time. The work sits at the intersection of four disciplines: environmental chemistry (what was released and in what chemical form), atmospheric science (where it went), exposure science (who contacted it, at what concentration, and for how long), and toxicology and risk assessment (what that dose implies for cancer and non-cancer endpoints).

The practical output is a defensible numerical estimate — micrograms per cubic meter at a given address in a given year, milligrams per kilogram per day of intake, an incremental lifetime cancer risk, a hazard quotient — accompanied by a transparent audit trail from the underlying record to the final number. The record is typically a mixture of permit applications, emission inventories, process descriptions, production volumes, stack tests, historical aerial photographs, Sanborn maps, deposition testimony, plant logs, and regulatory correspondence. The modeling converts that archival material into quantitative dose.

Three commitments govern the work. First, the methods are the published, agency-endorsed methods — U.S. EPA’s Guideline on Air Quality Models (Appendix W to 40 C.F.R. pt. 51) for dispersion, and EPA’s Risk Assessment Guidance for Superfund (RAGS) framework for dose and risk (U.S. EPA, 1989, 2009, 2017). Second, every assumption is stated, sourced, and tested — where an input is uncertain, the uncertainty is bounded rather than buried. Third, the analysis is designed to be reproducible by an opposing expert working from the same files: model input and output decks, meteorological data sets, emission calculations, and spreadsheets are retained and produced.

RoseyFields undertakes both affirmative and rebuttal assignments, and works for plaintiffs, defendants, municipalities, and state agencies. Methodologies are evaluated under Daubert and analogous standards.

2. Credentials and Years of Experience

Dr. Rosenfeld holds a Ph.D. in Soil Chemistry from the University of Washington (1999), with a dissertation on volatile organic compound filtration and the characterization, quantification, and control of odor emissions from biosolids applied to forest soil; an M.S. in Environmental Science from the University of California, Berkeley (1995); and a B.A. in Environmental Studies from the University of California, Santa Barbara (1991). He has more than 25 years of professional experience in exposure assessment, historical dose reconstruction, environmental forensics, and air dispersion modeling, spanning hundreds of contaminated sites nationwide.

He is founder and principal of RoseyFields LLC (2025–present) in Santa Monica, California, and was previously a founding partner and principal of Soil Water Air Protection Enterprise (SWAPE) from 2003 to 2025. Earlier positions include Senior Remediation Scientist at Komex H2O Science, Remediation Project Manager at Anteon Corporation and at Ogden (now Amec), and Risk Assessor at Bechtel. As a Remedial Project Manager under the U.S. Navy’s BRAC program at Treasure Island, he directed asbestos remediation in soils and friable building materials and conducted air monitoring to protect workers and adjacent communities. He is trained and certified under OSHA’s Hazardous Waste Operations and Emergency Response standard, 29 C.F.R. § 1910.120 (HAZWOPER).

His academic appointments include Lecturer and Assistant Researcher at the UCLA School of Public Health (2007–2011) and Adjunct Professor there (2003–2006), where he taught Environmental Health Science 100 — environmental health and epidemiology — to undergraduates, physicians, public health professionals, and nurses. He has also been an adjunct professor at San Diego State University and a lecturer for the National Ground Water Association.

Dr. Rosenfeld is co-author of The Risks of Hazardous Waste (Rosenfeld & Feng, 2011) and of three volumes of the Handbook of Pollution Prevention and Cleaner Production covering the petroleum, wood and paper, and agrochemical industries (Cheremisinoff & Rosenfeld, 2009, 2010, 2011). His peer-reviewed research includes cancer risk assessment for Camp Lejeune drinking water contamination from 1955 to 1987 (Rosenfeld et al., 2024), perfluoroalkyl substance exposure in firefighters (Rosenfeld et al., 2023), cancer risk and diesel exhaust exposure among railroad workers (Rosenfeld et al., 2022), hospital and community health burden after refinery fires in Richmond, California (Remy et al., 2019), and AERMOD-based evaluation of sulfur dioxide and VOC exposure in a Texas Gulf Coast community (Chen et al., 2012). He has published approximately forty peer-reviewed papers, book chapters, and books and delivered more than fifty conference presentations and invited lectures.

Testimony experience. Dr. Rosenfeld’s curriculum vitae, current as of July 20, 2026, itemizes 114 listed testimony entries across more than 100 distinct matters in state and federal courts in over 30 states, comprising approximately 120 deposition appearances, 16 trial appearances, and testimony at a contested Daubert hearing. The earliest listed matter dates to 2010 and the most recent to July 2026. Courts have included the United States District Courts for Vermont, Colorado, Nebraska, Hawaii, Wisconsin, Missouri, Ohio, Indiana, New York, Texas, Louisiana, Mississippi, Washington, Illinois, and New Jersey, and state trial and appellate courts in California, Florida, Texas, Illinois, Missouri, Ohio, Kentucky, Alabama, Georgia, Michigan, Virginia, Washington, Minnesota, Montana, Arizona, Iowa, Wisconsin, New York, Maryland, North Carolina, Nebraska, Oklahoma, and West Virginia.

Air dispersion modeling experience. Dr. Rosenfeld has performed or directed air dispersion modeling — principally AERMOD, together with AERSCREEN, ALOHA, CALPUFF, CALMET, BREEZE, and mass-balance box models — on approximately [N] projects over more than 25 years, spanning refineries and petrochemical plants, sterilization facilities, smelters and coke batteries, wood treatment operations, landfills and composting facilities, railyards and derailment sites, quarries and frac sand operations, and community-scale multi-source assessments.

A current curriculum vitae with the complete publication list and testimony history is available for download.

3. Model Selection: AERMOD, AERSCREEN, ALOHA, CALPUFF/CALMET, and Box Models

No single dispersion model fits every problem. Model selection is itself a methodological decision that must be justified, and Appendix W directs that the choice be driven by source type, averaging period, terrain, spatial scale, and the physics of the release (U.S. EPA, 2017).

AERMOD. AERMOD is EPA’s preferred near-field regulatory model for stationary sources out to approximately 50 kilometers. It was developed jointly by EPA and the American Meteorological Society and replaced the older ISC3 Gaussian framework. Its principal advances are planetary-boundary-layer similarity scaling rather than discrete Pasquill-Gifford stability classes; a bi-Gaussian probability density function for the convective boundary layer that reproduces the skewed vertical concentration distribution observed in unstable conditions; a dividing-streamline treatment of elevated terrain in which the plume is partitioned between a terrain-following and a horizontal-impacting state; and integration of the PRIME algorithms for building downwash and cavity entrainment (Cimorelli et al., 2005; Schulman et al., 2000; U.S. EPA, 2004a). AERMOD is used with three preprocessors: AERMET for meteorology, AERMAP for terrain and receptor elevations and hill-height scales, and AERSURFACE for surface characteristics. It accepts point, area, volume, line, and open-pit source types; supports hourly-varying emissions; and computes dry and wet deposition of particles and gases. For historical reconstruction its most important features are the ability to run long meteorological records year by year and to output period-average concentrations at fixed receptors — precisely the quantity needed for chronic dose.

AERSCREEN. AERSCREEN is the screening implementation of AERMOD (U.S. EPA, 2011a). It substitutes a matrix of synthetic worst-case meteorology for site-specific data and returns the maximum one-hour concentration as a function of downwind distance, with scaling factors for longer averaging periods. Its proper role is bounding: to determine whether a source could plausibly have produced concentrations of interest at a receptor, to rank multiple sources for further study, or to test whether a refined analysis is even warranted. Because it is deliberately conservative by design, AERSCREEN output is not an estimate of what actually occurred and should not be offered as one. Where a screening result is far below any level of concern, however, that finding is itself robust and can appropriately end an inquiry.

ALOHA. ALOHA, distributed by NOAA and EPA within the CAMEO suite, is the appropriate tool for short-duration accidental releases — tank ruptures, pipeline failures, derailments, valve failures, fires, and vapor cloud events (NOAA & U.S. EPA, 2013). It models both neutrally buoyant and dense (heavier-than-air) gas behavior, the latter using Britter-McQuaid-type relationships that AERMOD cannot represent, and it includes source-strength submodels for evaporating pools, pressurized releases, and boiling liquids. Its outputs are threat zones on a timescale of minutes to an hour, which map to acute exposure benchmarks such as AEGLs and ERPGs. ALOHA is not suited to chronic exposure or to complex terrain, and where an acute release must be evaluated against terrain or over longer periods, AERMOD with hourly-varying emissions is generally the more appropriate tool.

Box and mass-balance models. Box models solve a conservation-of-mass equation for a defined volume — a room, a shop bay, a valley, an urban airshed, a rail car interior, a locomotive cab — using a source term, a ventilation or air-exchange rate, and optionally a decay or deposition sink (Hanna et al., 1982). They are indispensable where the exposure occurred in a confined or semi-confined space, where the receptor was inside the source rather than downwind of it, or where the governing variable is air changes per hour rather than atmospheric stability. Related mass-balance and emission-estimation tools include EPA’s WATER9 and related air emission models for wastewater and waste management units, which estimate volatilization from impoundments, tanks, and treatment trains (U.S. EPA, 1994). Box models are simple enough to be checked by hand, which is an advantage in litigation, but they assume instantaneous and complete mixing, and that assumption must be examined rather than assumed.

CALPUFF and CALMET. CALPUFF is a non-steady-state Lagrangian puff dispersion model capable of simulating time- and space-varying meteorological conditions, calm and stagnant wind periods, long-range transport, and complex terrain effects that AERMOD’s steady-state plume formulation does not resolve. It is paired with CALMET, a diagnostic meteorological model that develops three-dimensional, time-varying wind and temperature fields from surface, upper-air, and terrain data, including terrain-driven flows such as slope and drainage winds. Under Appendix W, CALPUFF is treated as an alternative model rather than a preferred model for near-field, steady-state applications, and its use is documented and justified against the specific conditions — distance, terrain complexity, meteorological variability, or calm-wind frequency — that make AERMOD’s assumptions inappropriate for the problem at hand (U.S. EPA, 2017). CALPUFF has been evaluated against AERMOD in complex-terrain particulate studies, with comparative performance depending on site conditions and receptor distance (Tartakovsky et al., 2013).

Holmes and Morawska (2006) provide a useful review of the model classes and of the conditions under which each is appropriate.

4. Historical Emissions Reconstruction

Historical exposure reconstruction is the discipline of quantifying releases that occurred before anyone was measuring them. The methodological premise — well established in radiological dose reconstruction and directly transferable to chemical exposures — is that a defensible retrospective estimate can be built from process knowledge, production records, and emission factors, provided the reconstruction is explicit about its uncertainty (National Research Council, 1995; Till & Grogan, 2008).

The reconstruction typically proceeds in five steps.

Establish the operating history. Aerial photographs, Sanborn fire insurance maps, city directories, plant layouts, permit files, and construction and demolition records establish which units operated where, and when. Photointerpretation across a sequence of years frequently identifies stacks, vents, tanks, lagoons, waste piles, and rail spurs that appear in no document.

Reconstruct throughput. Production volumes, raw material purchase records, shipping manifests, batch sheets, and material safety data sheets establish the mass of chemical handled per unit time. Where records are incomplete, throughput is bracketed using rated capacity, employment figures, or industry-average utilization.

Apply emission factors. Uncontrolled and controlled emission factors from EPA’s AP-42 compilation, supplemented by stack test data, Toxics Release Inventory filings, National Emissions Inventory records, and the technical literature, convert throughput to mass emission rate (U.S. EPA, 1995a). Emission factor selection is documented to the specific AP-42 section, table, and rating, and the effect of control device installation dates is applied on the correct timeline — a distinction that frequently drives the difference between competing reconstructions.

Resolve the time series. Emissions are resolved to the finest defensible temporal scale: annual for chronic dose, and hourly or event-specific where shift patterns, seasonal campaigns, upsets, flaring, or startup and shutdown events matter.

Model and validate. The reconstructed emission series is run through the appropriate dispersion model, and predictions are compared against any contemporaneous measurement that exists — ambient monitors, complaint records, soil or attic dust concentrations, biological monitoring data, or vegetation surveys — as an independent check on the reconstruction. Multi-decade reconstructions of contaminated drinking water and of occupational diesel exhaust exposure illustrate the approach in published form (Rosenfeld et al., 2022, 2024).

5. Source Characterization

Dispersion model output is only as good as the source term, and source characterization is where most of the technical disagreement in air modeling litigation actually resides.

Each release is classified by geometry — point (a stack or vent), area (a lagoon, landfill surface, land application field, or spill footprint), volume (a building with fugitive leakage, or a windblown pile), line (a roadway or rail corridor), or open pit (a quarry or excavation). The classification determines the initial dispersion parameters and cannot be assigned casually; treating a fugitive building release as a point source, or a rail corridor as a single stack, produces errors in near-field concentration that can span an order of magnitude.

Point sources require stack height, inner diameter, exit gas temperature, and exit velocity — the latter two governing plume rise through buoyancy and momentum flux. Where records are missing, these parameters are reconstructed from engineering drawings, fan and blower specifications, combustion calculations, or photographic scaling. Building downwash is evaluated using the PRIME algorithms with building dimensions derived from drawings or photogrammetry, and is critical wherever stack height is less than roughly 2.5 times adjacent building height (Schulman et al., 2000). Area and volume sources require release height, initial vertical dimension, and a flux estimate; for open liquid surfaces this generally requires a volatilization calculation using compound-specific Henry’s law constants and mass transfer coefficients (U.S. EPA, 1994).

Fugitive emissions — leaking valves, flanges, pumps, seals, loading racks, open doors, transfer points, and windblown dust — are estimated from component counts and leak-rate factors, from tracer or measurement studies, or by mass balance closure against inventory. In many industrial reconstructions fugitives dominate near-field ground-level concentration even where stack emissions dominate total mass, because fugitives are released at low elevation with little plume rise. Failing to characterize them is a common and consequential omission.

Where multiple facilities and mobile sources contribute, each is modeled separately so that source apportionment at each receptor can be reported explicitly. This separation is what allows the analysis to answer the attribution question — how much of the concentration at a given address came from which source — rather than only the aggregate exposure question.

6. Meteorological Data

Meteorology is the transfer function between emissions and exposure, and for historical reconstruction it is often the single most demanding data requirement.

AERMOD requires paired hourly surface and upper-air observations processed through AERMET, which computes the boundary-layer parameters the model actually uses: friction velocity, Monin-Obukhov length, surface heat flux, convective velocity scale, and mechanical and convective mixing heights (U.S. EPA, 2023). Surface data are typically drawn from National Weather Service Automated Surface Observing System stations or from site-specific towers; upper-air soundings come from the nearest radiosonde site. Selection of a representative station is a documented judgment based on distance, terrain similarity, surface roughness, coastal or valley influence, and comparability of the wind rose — not simply on proximity.

Surface characteristics — albedo, Bowen ratio, and surface roughness length — are assigned by AERSURFACE from land cover data, by direction sector and season, using the land cover that existed during the exposure period rather than current conditions. Historical land cover is reconstructed from aerial photography and historical land use data where the landscape has changed materially, as it usually has over a multi-decade exposure window.

Appendix W generally calls for five consecutive years of representative National Weather Service data, or at least one year of site-specific data (U.S. EPA, 2017). For a fifty-year reconstruction, hourly data will not exist for the earliest years. The defensible approach is to model the years for which data exist, examine interannual variability across those years, and either apply a representative multi-year average with a stated variability range or, where a prognostic reanalysis is appropriate, generate meteorology from WRF output through the Mesoscale Model Interface (MMIF). Whichever route is taken, the limitation is stated in the opinion rather than concealed within it. Kumar, Dikshit, and Patil (2021) found improved model performance using WRF-generated meteorology relative to observed station data in an industrial region, and Rzeszutek and colleagues have documented the sensitivity of AERMOD output to meteorological input choices generally.

Calm and light-wind hours warrant particular attention. AERMOD’s steady-state formulation does not resolve plume meander well at very low wind speeds, and the treatment of calms — which AERMOD excludes from averaging — can bias long-term averages in low-wind regimes. Where calms are frequent, sensitivity analysis on the calm treatment is appropriate.

7. Receptors and Exposure Areas

A receptor is a point in space where concentration is calculated, and the receptor design determines what the analysis can and cannot say.

Discrete receptors are placed at the locations that matter to the question presented: residential addresses, schools, daycare centers, hospitals, workplaces, parks, athletic fields, and — where individual plaintiffs are at issue — each residence of record, with flagpole receptors at appropriate breathing heights for multi-story structures. Cartesian or polar grids, typically nested with fine spacing near the source and coarser spacing at distance, establish the concentration field and locate the maximum. Fenceline receptor rings support comparison with any historical perimeter monitoring. Receptor elevations and hill-height scales are assigned by AERMAP from digital elevation data.

The output at each receptor is post-processed into the averaging periods that match the relevant toxicity benchmark: one-hour and 24-hour maxima for acute comparison against AEGLs, ERPGs, or reference concentrations for short-term exposure; annual and period averages for chronic inhalation dose. Where residential history is known, receptor-specific annual averages are assembled into a year-by-year exposure profile for each individual, so that the exposure duration used in the dose calculation reflects actual residence rather than a default.

Isopleth mapping of modeled concentration in a geographic information system produces the exposure area or exposure zone — the geographic delineation of where a defined concentration or dose was exceeded. Courts have relied on AERMOD-based geographic delineation of exposure zones from refinery emissions, and the method is standard in community exposure assessment (Chen et al., 2012). Overlaying isopleths on parcel and census data yields population counts within each exposure band. Where indoor exposure is at issue, modeled outdoor concentrations are converted using indoor/outdoor ratios or infiltration modeling, and where activity patterns matter, time-weighted microenvironmental exposure is constructed using time-activity data from EPA’s Exposure Factors Handbook (U.S. EPA, 2011b). Isakov et al. (2009) describe the general architecture of linking dispersion output to exposure models for health studies.

8. Deposition and Multimedia Pathways

Inhalation is rarely the whole exposure. Particles and gases deposit, and deposition creates secondary pathways that can dominate total dose for persistent, bioaccumulative compounds — dioxins and furans, PCBs, PAHs, metals, and PFAS among them.

AERMOD computes dry deposition as the product of modeled air concentration and a deposition velocity that accounts for particle size distribution and density, gravitational settling, aerodynamic and quasi-laminar resistance, atmospheric stability, and surface characteristics; wet deposition is computed from precipitation rate, airborne concentration, and compound- and phase-specific scavenging coefficients (U.S. EPA, 2004b). Because deposition velocity is strongly dependent on particle size, the particle size distribution and the mass fraction in each size bin are among the most sensitive inputs in any deposition analysis, and they must be sourced to stack test data or to the literature for the specific process — not assumed.

Modeled deposition flux, in mass per unit area per unit time, is the entry point to multimedia modeling. EPA’s Human Health Risk Assessment Protocol for Hazardous Waste Combustion Facilities provides the standard equation set for propagating deposition into soil, produce, forage, animal tissue, surface water, sediment, and fish, and for calculating the resulting ingestion doses (U.S. EPA, 2005c). The soil pathway alone supports incidental soil ingestion, dermal contact, and inhalation of resuspended dust; the home-produce and backyard-poultry-and-egg pathways frequently dominate dioxin dose in residential settings; and the water and fish pathways govern in watershed cases.

Deposition modeling also enables independent validation of the reconstruction, because deposited mass persists. Attic dust, undisturbed surface soil, tree bark, and sediment cores function as historical archives of past airborne concentration, and comparing measured concentrations in those media against modeled cumulative deposition provides a physical check on the emission reconstruction. Dr. Rosenfeld’s published work has used attic dust and paired human blood lipid measurements in exactly this way near wood treatment facilities and former PCB production and secondary copper operations (Hensley et al., 2007; Feng et al., 2010; Gonzalez et al., 2010). Published field validations of AERMOD deposition are discussed in Section 9.

9. Model Validation and Sensitivity Analysis

An opinion built on a model is only as strong as the evidence that the model performs, and the evidence for AERMOD is extensive.

Developmental and regulatory evaluation. AERMOD was subjected to independent scientific peer review during its promulgation and to a formal evaluation against seventeen field study databases spanning urban and rural settings, flat and complex terrain, elevated and surface releases, and sources with and without building downwash (Perry et al., 2005; U.S. EPA, 2002). The evaluation focused on the upper end of the concentration distribution — the region most relevant to regulatory and exposure applications — and concluded that with few exceptions AERMOD’s performance was superior to the other dispersion models tested. This remains one of the most comprehensive validation efforts performed for any atmospheric dispersion model, and it is the basis on which AERMOD was designated the preferred near-field regulatory model under Appendix W (U.S. EPA, 2017). Hanna et al. (2001) reached comparable conclusions in an independent evaluation against the OPTEX, Duke Forest, Kincaid, Indianapolis, and Lovett datasets.

Controlled tracer and near-road evaluation. Heist et al. (2013) evaluated AERMOD against tracer releases in which emission rates were precisely known — eliminating emission uncertainty — and matched predictions to observations by exact time and location, a stricter test than conventional distributional comparison. AERMOD predictions fell within a factor of two of measured concentrations 76 to 78 percent of the time in the highway study, performing comparably to ADMS and R-LINE and outperforming CALINE.

Industrial particulate and complex terrain. Tartakovsky, Broday, and Stern (2013) compared AERMOD and CALPUFF against measured TSP and PM₁₀ near quarries in complex terrain and found AERMOD in better agreement with measurements. Fadavi et al. (2016) modeled TSP from a cement facility against four monitoring locations and reported predicted-to-observed ratios of 0.949 to 1.456, with all modeled values within a factor of two. Omidvarborna et al. (2018) reported a predicted-to-observed ratio of 0.976 at one receptor near an iron processing facility. Gibson et al. (2013) evaluated AERMOD for PM, NOₓ, and SO₂ from point and major line sources, and Olesen et al. (2011) evaluated AERMOD/PRIME against wind tunnel data for stack-building configurations.

Deposition validation. Omidvarborna et al. (2018) compared modeled deposition against dust-fall, soil, and air measurements and reported a predicted-to-observed dust-fall ratio of 1.004. Barton, Zarzecki, and Russell (2010) provide the most direct validation for a perfluorinated compound, comparing a 24-hour PFOA air concentration simulation against nine perimeter monitors over nine sampling events and a one-year deposition simulation against 42 soil and grass samples at 16 locations near a fluoropolymer facility; AERMOD located the maximum concentration correctly 89 percent of the time, and modeled deposition agreed with measured soil and grass concentrations across most of the study area. Where predictions diverged, the model tended to overpredict — a tendency the authors characterized as a positive result, since conservative overprediction is preferable to underprediction when outputs inform human exposure assessment. Moradi et al. (2021) found measured chromium, lead, and zinc in soils and vegetation near a cement facility significantly correlated with AERMOD-predicted particulate patterns, and Zeng et al. (2023) found that incorporating AERMOD deposition into spatial interpolation improved prediction for five of eight metals near a coke plant and reduced root-mean-square error by as much as 27.5 percent.

Site-specific validation and sensitivity analysis. Published performance is necessary but not sufficient; each application requires its own verification. Where contemporaneous monitoring data exist, modeled and measured concentrations are compared using standard performance metrics — fractional bias, geometric mean bias, normalized mean square error, correlation, and factor-of-two agreement. Where they do not, the reconstruction is checked against deposited-media measurements, complaint or odor records, or biomonitoring data. Sensitivity analysis is then run across the inputs that plausibly vary: emission rate, control efficiency and control installation date, particle size distribution, stack parameters, meteorological year, surface roughness, calm treatment, receptor height, and building dimensions. The product is a range with identified drivers rather than a single unqualified number, which is both better science and more useful to a trier of fact. Known model limitations — steady-state assumption, low-wind performance, downwash algorithm constraints, and the near-field domain limit — are stated affirmatively.

10. Cancer and Non-Cancer Calculations

Modeled concentration becomes dose, and dose becomes risk, through EPA’s standard risk assessment architecture (U.S. EPA, 1989, 1991, 2009).

Inhalation. Current EPA practice for the inhalation route uses the exposure concentration rather than an intake rate, per RAGS Part F (U.S. EPA, 2009). The exposure concentration is the modeled air concentration adjusted for exposure time, frequency, and duration and averaged over the appropriate period. Cancer risk is the product of the lifetime-averaged exposure concentration and the inhalation unit risk (IUR); non-cancer hazard is the ratio of the averaging-period exposure concentration to the reference concentration (RfC). Toxicity values are taken from EPA’s Integrated Risk Information System, and where IRIS lacks a value, from the hierarchy of alternative sources — the Provisional Peer-Reviewed Toxicity Values, ATSDR minimal risk levels, and California OEHHA values — with the selection and its rationale disclosed.

Ingestion and dermal. For deposition-driven pathways, dose is calculated as average daily dose in milligrams per kilogram per day, combining medium concentration, intake rate, exposure frequency and duration, body weight, and averaging time. Cancer risk is dose multiplied by the oral slope factor; non-cancer hazard is dose divided by the reference dose. Dermal exposure follows RAGS Part E, with absorbed dose calculated from adherence factors, surface area, and compound-specific dermal absorption fractions (U.S. EPA, 2004c). Exposure parameters are drawn from the Exposure Factors Handbook and from RAGS standard default exposure factors, with age-specific values used for child receptors (U.S. EPA, 1991, 2011b).

Aggregation, susceptibility, and statistics. Pathway-specific risks are summed to a total incremental lifetime cancer risk and compared against the CERCLA 10⁻⁶ to 10⁻⁴ risk range; hazard quotients are summed into hazard indices, segregated by target organ and mechanism of action where the toxicology supports segregation (U.S. EPA, 1989; ATSDR, 2018, 2022). For chemicals acting through a mutagenic mode of action, age-dependent adjustment factors are applied to early-life exposures per EPA’s supplemental carcinogen guidance (U.S. EPA, 2005a, 2005b). Where the exposure point concentration is derived from measured data, the 95 percent upper confidence limit on the mean is used, computed with ProUCL using the distribution-appropriate estimator, consistent with EPA guidance on calculating the concentration term (U.S. EPA, 1992, 2002b, 2015).

Acute assessment. For short-duration releases, modeled one-hour and short-term maxima are compared against AEGL-1, -2, and -3 values, ERPGs, and acute reference exposure levels, and the duration of exceedance and the area affected are reported.

Uncertainty. Every calculation is accompanied by a quantitative or qualitative uncertainty analysis identifying the dominant contributors — typically emission rate, toxicity value, exposure duration, and the choice of central-tendency versus reasonable-maximum exposure. Where the data support it, probabilistic analysis under RAGS Volume III replaces point estimates with distributions (U.S. EPA, 2001).

11. Daubert and Litigation Considerations

Federal Rule of Evidence 702, as amended effective December 1, 2023, requires the proponent of expert testimony to demonstrate more likely than not that the testimony rests on sufficient facts or data, is the product of reliable principles and methods, and reflects a reliable application of those principles and methods to the facts of the case. Daubert v. Merrell Dow Pharmaceuticals (1993), General Electric Co. v. Joiner (1997), and Kumho Tire Co. v. Carmichael (1999) supply the governing reliability framework, and the Federal Judicial Center and National Research Council’s Reference Manual on Scientific Evidence (2011) provides the standard judicial reference for evaluating exposure and dispersion evidence. Many states apply Daubert or an analogous standard; others retain a Frye general-acceptance test. Methodologies are evaluated under Daubert and analogous standards, and the design of the analysis anticipates that scrutiny from the outset.

Air dispersion modeling maps onto the reliability factors unusually well. The methods are published and peer reviewed in the primary literature (Cimorelli et al., 2005; Perry et al., 2005). They have known and quantified error rates, documented across dozens of field validation studies (Section 9). They are governed by controlling standards — Appendix W for dispersion, RAGS for dose and risk — and those standards are promulgated by the federal agency with regulatory jurisdiction. They were developed for regulatory and scientific purposes wholly independent of litigation, and they are testable and reproducible: the input decks, meteorological files, and output files can be re-run by any opposing expert.

The reliability problems in practice therefore rarely concern the models themselves. They concern the application: source terms not tied to record evidence, emission factors selected without documented basis, control-device timelines misaligned with the exposure period, receptor placement that does not correspond to the plaintiffs’ actual residences, screening-model output presented as a reconstruction of what happened, averaging periods mismatched to the toxicity benchmark, missing sensitivity analysis, unstated calms or meteorological substitutions, and — most commonly — an analytical gap between the modeled concentration and the specific exposure or causation opinion the modeling is offered to support. Each of these is addressed on the front end: the analysis is scoped so that the record supports each input, the fit between the modeled quantity and the legal question is stated explicitly, and limitations are disclosed rather than defended after the fact.

Rebuttal work applies the same framework in reverse, examining an opposing model’s source characterization, meteorological representativeness, receptor design, averaging-period selection, and QA documentation, and — where the input files have been produced — re-running the model to test the sensitivity of its conclusions to its assumptions.

12. Representative Contaminants and Facilities

Contaminants. Benzene, toluene, ethylbenzene, and xylenes; 1,3-butadiene; ethylene oxide; vinyl chloride; trichloroethylene, tetrachloroethylene, methylene chloride, and other chlorinated solvents; 1,4-dioxane; 1,2,3-trichloropropane; hydrogen sulfide, sulfur dioxide, nitrogen oxides, carbon monoxide, and carbon dioxide; ammonia and reduced sulfur compounds; polycyclic aromatic hydrocarbons and creosote constituents; pentachlorophenol; polychlorinated biphenyls; 2,3,7,8-TCDD and other polychlorinated dibenzodioxins and dibenzofurans; per- and polyfluoroalkyl substances including PFOA and PFOS; diesel particulate matter and diesel exhaust organics; respirable crystalline silica; asbestos; lead, arsenic, cadmium, chromium including hexavalent chromium, mercury, manganese, nickel, and zinc; PM₁₀ and PM₂.₅; pesticides and herbicides including atrazine, glyphosate, and chlorinated organics; brominated flame retardants; microplastics and engineered nanoparticles; and landfill gas and compost odorants.

Facilities and source types. Petroleum refineries and petrochemical plants, including fire, flaring, and upset events; chemical manufacturing complexes; medical device and spice sterilization facilities using ethylene oxide; coke batteries and steel operations; secondary smelters and lead recycling operations; wood treatment and creosoting plants; pulp, paper, and wood pellet facilities; cement plants and aggregate operations; foundries and metal finishing and plating shops; municipal and hazardous waste landfills, transfer stations, and incinerators; composting facilities, biosolids land application sites, and anaerobic digesters; wastewater treatment plants; railyards, rail corridors, locomotive shops, and derailment sites; military installations, firing ranges, and airports using aqueous film-forming foam; frac sand mining and processing; oil and gas well pads and hydraulic fracturing operations; countertop and stone fabrication shops; power plants; dry cleaners; and warehouse and industrial fires.

13. Five Short Case Examples

Consistent with RoseyFields’ confidentiality practice, the following are described without client, counsel, facility, or location identifiers.

Refinery release and community exposure zone. Following a release event at a Gulf Coast petroleum refinery, hourly emission rates were reconstructed from process data, flare and vent records, and event chronology, then modeled in AERMOD with site-specific meteorology and a fine receptor grid across the surrounding residential area. Modeled short-term concentrations of sulfur dioxide and volatile organics were compared against contemporaneous ambient measurements, and isopleths were used to delineate the geographic zone within which defined concentration thresholds were exceeded — supporting both exposure-zone definition and per-address exposure estimates. The general methodology is documented in the peer-reviewed literature (Chen et al., 2012).

Multi-decade railroad diesel exposure reconstruction. For occupational diesel exhaust exposure among rail workers, exposure was reconstructed by combining locomotive and shop equipment inventories, duty cycles, fuel consumption, ventilation conditions, and published emission factors, using enclosed-space mass-balance calculations for cab and shop environments and dispersion modeling for yard-wide exposure. Cumulative dose was assembled year by year from job history and compared against the epidemiological literature on diesel exhaust and cancer (Rosenfeld et al., 2022).

Ethylene oxide sterilization facility. Emissions from a commercial sterilization operation were reconstructed from sterilizer cycle counts, ethylene oxide purchase and usage records, control device installation dates and efficiencies, and permit files, resolved to an annual time series across the facility’s operating history. AERMOD was run with representative multi-year meteorology and receptors at residences, schools, and daycare facilities; lifetime-averaged exposure concentrations were converted to incremental cancer risk using the current inhalation unit risk, with age-dependent adjustment factors applied to early-life exposure.

Wood treatment facility deposition and multimedia dose. For a former wood treating operation, dioxin, furan, and PAH emissions were reconstructed and modeled for both concentration and deposition. Modeled cumulative deposition was compared against measured attic dust and surface soil concentrations as an independent physical check on the emission reconstruction, and deposition flux was propagated through soil, home produce, and incidental ingestion pathways to total dose. Paired attic dust and human blood lipid data from facilities of this type are reported in the peer-reviewed literature (Hensley et al., 2007; Feng et al., 2010).

PFAS air deposition to soil and groundwater. For a fluoropolymer and PFAS-handling facility, stack and fugitive PFOA emissions were estimated from process and inventory records and modeled in AERMOD for both ambient concentration and wet and dry deposition. Modeled deposition patterns were compared against measured soil concentrations and against the distribution of impacted private wells, supporting an air-deposition pathway to soil and groundwater distinct from direct discharge. The validity of AERMOD for perfluorinated compound dispersion and deposition is established in the published literature (Barton et al., 2010).

14. Ten Frequently Asked Questions

1. Can exposure be quantified when no one was monitoring at the time? Yes, within stated uncertainty. Historical exposure reconstruction is an established discipline with a substantial methodological literature (National Research Council, 1995; Till & Grogan, 2008). Emissions are reconstructed from process, production, and regulatory records using published emission factors, then modeled with representative meteorology. What must accompany the estimate is an honest uncertainty analysis and, wherever possible, validation against some independent physical measurement.

2. How accurate is AERMOD? Across field validation studies AERMOD predictions commonly fall within a factor of two of measured concentrations — 76 to 78 percent of the time in a controlled tracer study where emission rates were exactly known (Heist et al., 2013) — with predicted-to-observed ratios near unity in several industrial particulate evaluations (Fadavi et al., 2016; Omidvarborna et al., 2018). Factor-of-two agreement is the recognized performance benchmark for atmospheric dispersion models, and AERMOD generally meets or exceeds it. Accuracy in any specific application depends more on the quality of the source term than on the model.

3. Does AERMOD over- or under-predict? It does both, depending on source and site conditions, with no evidence of systematic failure. The validation record includes ratios both above and below unity (Fadavi et al., 2016; Omidvarborna et al., 2018; Tartakovsky et al., 2013). In the PFOA evaluation the tendency was toward conservative overprediction, attributable to the steady-state assumption of constant hourly wind direction, which does not capture short-term plume meander (Barton et al., 2010).

4. How much meteorological data is required, and what if it does not exist for the exposure period? Appendix W generally calls for five consecutive years of representative National Weather Service data or one year of site-specific data (U.S. EPA, 2017). For older exposure periods, the practice is to model the years for which data exist, quantify interannual variability, and state the representativeness assumption explicitly; prognostic meteorology generated from WRF through MMIF is an alternative where appropriate (Kumar et al., 2021).

5. When is a screening model appropriate, and when is it not? AERSCREEN is appropriate for bounding, for ranking sources, and for demonstrating that concentrations could not plausibly have reached a level of concern (U.S. EPA, 2011a). It is not appropriate as an estimate of concentrations that actually occurred, because it substitutes worst-case synthetic meteorology for real conditions. Presenting screening output as a reconstruction is a recurring methodological error.

6. Why use ALOHA instead of AERMOD for a derailment or tank failure? ALOHA models dense-gas behavior and includes source-strength submodels for evaporating pools, pressurized releases, and boiling liquids that AERMOD lacks, and it operates on the minutes-to-hours timescale relevant to acute exposure (NOAA & U.S. EPA, 2013). Where terrain, longer duration, or deposition matter, AERMOD with hourly-varying emissions is generally the better tool, and the two are sometimes used together.

7. Is modeling admissible, or is measured data required? Air dispersion modeling is routinely admitted. It is peer reviewed, has known error rates, is governed by promulgated federal standards, was developed independent of litigation, and is testable and reproducible — the classic reliability factors under Rule 702 and Daubert. Measured data, where available, strengthens a modeling opinion by permitting validation, but the absence of contemporaneous measurement does not by itself render modeling unreliable. Methodologies are evaluated under Daubert and analogous standards.

8. How is exposure from multiple facilities separated? Each source is modeled independently, and receptor-specific results are reported by source. This produces explicit apportionment at each location — what fraction of concentration came from which facility — which is what attribution requires. Aggregate modeling alone cannot answer the attribution question.

9. What controls total dose when a contaminant deposits? For persistent, bioaccumulative compounds, ingestion pathways driven by deposition often exceed direct inhalation. Deposition flux is propagated into soil, produce, forage, animal products, water, and fish using EPA’s standard multipathway equation set (U.S. EPA, 2005c). Deposition also enables validation, because deposited mass persists in attic dust, undisturbed soil, and sediment and can be measured today as a record of past airborne concentration.

10. What is needed to begin, and what is produced? Useful starting materials include permit and air files, emission inventories and stack tests, process and production records, plant layouts and engineering drawings, aerial photographs, any ambient or perimeter monitoring data, soil or dust sampling results, and residential or work history for the individuals at issue. The deliverable is a report documenting source characterization, model configuration and justification, meteorological basis, receptor design, validation, sensitivity analysis, dose and risk calculations, and uncertainty — accompanied by the model input and output files, meteorological data sets, and calculation spreadsheets, so that the analysis can be independently reproduced.

References