{"slug":"how-accurate-is-aermod","title":"How Accurate Is AERMOD? Model Evaluation, Uncertainty and What It Means for Your Study","excerpt":"AERMOD is built to predict the high end of the concentration distribution, not the value at a given hour and place. Here is what evaluations show, where uncertainty really comes from in Indian studies, and how to test and report it.","date":"2026-11-12","author":"DSTechnoverse","image":"/blog-images/aermod-guides/how-accurate-is-aermod-paired-unpaired.svg","category":"Air Dispersion Modelling","tags":["AERMOD","Model Evaluation","Uncertainty","Sensitivity Analysis","EIA","Air Dispersion Modelling","NAAQS"],"content":"> **Quick answer:** AERMOD is designed to predict the high end of the concentration distribution, the highest values that occur somewhere in the study area at some time, rather than the exact concentration at a given place and hour. Evaluations against field data show it generally estimates peak concentrations within about a factor of two, and often much closer. In a real Indian study, the accuracy of the result depends more on the emission rates, meteorology and background data you feed it than on the model itself.\n\n\"How accurate is AERMOD?\" is one of the most common questions we hear from project proponents and EIA reviewers, usually asked after a contour map shows a maximum ground level concentration (GLC) of, say, 4.8 µg/m³ and someone wonders whether that number can be trusted to one decimal place. The honest answer is that it cannot, but it does not need to be. A regulatory dispersion model is a screening and decision tool. It is built to answer one question reliably: will this source, under the full range of weather it is likely to see, push concentrations in the surrounding area above an acceptable level?\n\nThis guide explains what accuracy means for AERMOD, how it was evaluated, where the real uncertainty sits in Indian studies, and how to test and communicate it in an EIA or Consent to Establish (CTE) report. If you are new to the model itself, start with our explainer on [how AERMOD works](/knowledge-base/how-aermod-works/).\n\n## What Accuracy Means for a Regulatory Model\n\nWhen most people think of accuracy, they imagine a forecast: the model says 42 µg/m³ at the village school at 9 am tomorrow, and a monitor records 40 µg/m³. That kind of agreement, paired in time and space, is not what AERMOD is designed to deliver, and no steady-state Gaussian or near-Gaussian model reliably delivers it.\n\nThe reason is physical. A shift in wind direction of five or ten degrees moves the plume centreline hundreds of metres sideways at a receptor 3 km away, and the hourly wind in a met file is an average from one sensor that may be kilometres from the stack. So in any given hour the model may put the plume over the school while the real plume passed over the next field. Weak paired correlation is expected.\n\nWhat regulators actually need is different. They need to know the highest concentration likely to occur anywhere in the area of interest, over the averaging period that matters for the standard. For Indian NAAQS 2009 that means, for example, the 24-hour PM10 standard of 100 µg/m³ and the annual PM10 standard of 60 µg/m³, or the 24-hour SO2 and NO2 standards of 80 µg/m³. A model is fit for this purpose if the top of its predicted concentration distribution matches the top of the observed distribution, regardless of exactly when and where each value occurs.\n\nThe US EPA Guideline on Air Quality Models (40 CFR Part 51, Appendix W), which is the main reference for AERMOD's regulatory use, makes this point directly. It notes that models are more reliable for longer averaging times than for short-term concentrations at specific locations, and that they are reasonably reliable in estimating the magnitude of the highest concentrations occurring sometime, somewhere within an area. It describes errors in highest estimated concentrations of roughly 10 to 40 percent as typical, comfortably within the often-quoted factor of two, while warning that estimates for a specific time and site are poorly correlated with observations.\n\n| Type of comparison | What it asks | How AERMOD typically performs | Relevance to EIA |\n|---|---|---|---|\n| Paired in time and space | Does the model match the monitor at this receptor in this hour? | Weak; scatter plots show wide spread | Low, not the regulatory question |\n| Paired in space, unpaired in time | Does the highest value at this receptor match, whatever the hour? | Moderate | Useful for checking a specific monitor |\n| Unpaired in time and space | Does the top of the predicted distribution match the top of the observed one? | Generally good, often within a factor of two | High, this is what compliance depends on |\n| Long-term averages | Does the annual mean field look right? | Generally good where emissions and meteorology are representative | High for annual NAAQS |\n\nSo the right question is not \"how accurate is AERMOD at a point?\" but \"how well does AERMOD reproduce the high end of the concentration distribution, and how sensitive is my result to the inputs I have chosen?\"\n\n![Paired versus unpaired comparison of model and observed concentrations, showing weak hourly agreement but good agreement in the top ranked values](/blog-images/aermod-guides/how-accurate-is-aermod-paired-unpaired.svg)\n\n## How the US EPA Evaluated AERMOD\n\nAERMOD was developed by the AERMIC committee (American Meteorological Society and US EPA) to replace ISC3, and became the preferred near-field regulatory model in the 2005 revision of Appendix W after an extensive performance evaluation documented in EPA reports and peer-reviewed papers.\n\nDescribed in general terms, the evaluation used a set of field studies that covered the situations AERMOD is meant to handle:\n\n1. **Short-range tracer studies over flat terrain.** Experiments such as Prairie Grass released a tracer gas near the ground and measured it on arcs of samplers downwind. These test the near-field dispersion formulation for low-level releases.\n2. **Tall stacks in flat or rolling terrain.** Long-term SO2 monitoring networks around power plants, such as the Kincaid study, test how the model handles buoyant plumes from tall stacks in convective and stable conditions.\n3. **Complex terrain.** Datasets from plants located near hills and ridges, such as Lovett, test the terrain formulation that splits the plume between a horizontal state and a terrain-following state.\n4. **Building downwash.** Wind-tunnel and field data for stacks affected by nearby structures were used when the PRIME downwash algorithm was incorporated.\n5. **Urban settings.** Studies with SF6 tracers released in a city test the urban boundary layer treatment.\n\nThe developers compared AERMOD against observations and against the models it was meant to replace. The headline finding, consistently reported, was that AERMOD reproduced the upper end of the observed concentration distribution well across these settings, and generally better than ISC3 and the older complex terrain models. In some datasets it overpredicted modestly and in others it underpredicted modestly, but it did not show the large systematic biases seen in some earlier models.\n\nOne caveat matters for Indian users: the evaluation datasets are North American, with well-characterised emissions and on-site meteorology. The model has been revised many times since, with further evaluation, and the conclusion has held: when inputs are good, the model is reliable for regulatory purposes.\n\n## The Factor of Two and Other Performance Statistics\n\nThe claim that regulatory models are \"accurate within a factor of two\" is shorthand. Here is what lies behind it.\n\n### Factor of Two (FAC2)\n\nFAC2 is the fraction of predictions between half and double the corresponding observations. For paired hourly data it can be low for any model. For the highest concentrations compared unpaired, a well-performing model lands within the band, meaning the predicted peak is likely between half and twice the true peak, and often much closer.\n\n### Fractional Bias (FB)\n\nFractional bias compares the mean (or another summary such as the RHC) of predictions and observations, normalised so that it runs from minus 2 to plus 2. It is calculated as twice the difference between observed and predicted, divided by their sum. Depending on the sign convention used, a positive or negative value indicates over or underprediction, so always check how the study defines it. An FB of zero is perfect agreement, and an FB of about 0.67 in magnitude corresponds to a factor of two. FB is popular because it treats overprediction and underprediction symmetrically.\n\n### Robust Highest Concentration (RHC)\n\nThe single highest observed or predicted value is a noisy statistic: one unusual hour can dominate it. The robust highest concentration, introduced by Cox and Tikvart for EPA model evaluation, smooths this by fitting an exponential tail to the top few values (commonly the top 26). The RHC is calculated from the nth highest value and the average of the values above it, and it gives a stable estimate of the peak that is less affected by a single outlier. Comparing predicted and observed RHC, and expressing the result as a fractional bias, is the core of the EPA's operational evaluation protocol.\n\n### Quantile-Quantile (Q-Q) Plots\n\nA Q-Q plot ranks predicted values from highest to lowest, ranks observed values the same way, and plots them against each other. Rank 1 predicted against rank 1 observed, rank 2 against rank 2, and so on. If the points sit on the 1:1 line, the distributions match, even though the individual values occurred at different times. The factor-of-two lines are usually drawn either side. Q-Q plots are the most intuitive way to show regulatory performance, and they are the plot to ask for if anyone challenges a model with monitoring data.\n\n| Statistic | What it measures | Ideal value | Typical use |\n|---|---|---|---|\n| FAC2 | Share of predictions within 0.5 to 2 times observed | 1 (100%) | Overall agreement; useful for top-ranked values |\n| Fractional bias (FB) | Normalised difference in means or RHC | 0 | Over or underprediction tendency |\n| Robust highest concentration (RHC) | Stable estimate of the peak | Predicted equals observed | EPA operational evaluation |\n| Q-Q plot | Match of ranked distributions | Points on 1:1 line | Visual communication of regulatory performance |\n\nA typical Indian EIA will not include a formal evaluation, as the plant does not yet exist. These statistics still help when a reviewer asks how good the model is, and become directly useful for expansion projects with monitoring data, discussed below.\n\n## Where the Real Uncertainty Lies in Indian Studies\n\nThe evaluation literature tells us AERMOD's physics is sound. In practice, almost every large error we see in Indian reports comes from inputs and set-up, not from the model formulation. The table below ranks the main sources roughly by how much they usually matter.\n\n| Source of uncertainty | Typical effect on predicted peak | Main cause | How to reduce it |\n|---|---|---|---|\n| Emission rates | Directly proportional; a 2x error gives a 2x error | Using design maxima, wrong units, missing sources, outdated factors | Build a documented inventory with stack-wise basis |\n| Meteorology representativeness | Can shift peaks by a factor of two or more, and move their location | Distant station, coastal or valley effects, poor data capture | Use site data or a well-justified surrogate; check wind rose |\n| Low wind speeds and calms | Can understate or overstate stable-hour peaks | Many calm hours in Indian winters; sensor threshold | Review calm statistics; consider low wind options with justification |\n| Background concentrations | Often larger than the increment itself | Short campaigns, unrepresentative stations | Use the 98th percentile of a representative baseline |\n| Terrain and land use | Moderate to large near hills; small on flat plains | Flat-terrain runs near ridges; wrong surface parameters | Use terrain data where relevant; check surface roughness and albedo |\n| Building downwash | Can be large near short stacks | BPIP not run or buildings omitted | Run BPIP for stacks below good engineering practice height |\n\n### Emission Rates\n\nPredicted concentration is directly proportional to the emission rate, so a doubled emission rate doubles every increment. This is often the largest single source of error in Indian studies, and it is usually invisible in the final report. Common problems include using the CPCB or SPCB emission limit as the emission rate without saying so, converting mg/Nm³ to g/s with the wrong flue gas volume (actual instead of normal conditions, or dry instead of wet), forgetting fugitive and area sources, and missing secondary stacks such as DG sets, boilers or thermic fluid heaters.\n\nUsing the permitted emission limit for each stack is a defensible conservative choice for a CTE application, provided you state it. Our guide to [emission inventories for dispersion modelling](/blog/emission-inventory-for-dispersion-modelling/) covers the calculation steps, and [stack emission monitoring data analysis](/blog/stack-emission-monitoring-data-analysis/) explains how to use measured data for an existing plant.\n\n### Meteorology Representativeness\n\nAERMOD needs hourly surface data and upper air information, processed through AERMET. The question is not just whether the data is complete but whether it represents the site. A station 40 km away across a river valley, or on the coast when the plant is inland, can produce a wind rose that bears little resemblance to local conditions. Because the peak 24-hour concentration usually occurs in a particular combination of wind direction and stability, a misrepresented wind rose can move the location of the maximum, and change its value substantially.\n\nData capture matters too: careless gap filling or a sensor with a high starting threshold will change the result. The [AERMET preprocessing guide](/blog/aermet-meteorological-data-preprocessing/) and our article on [meteorological data for air modelling in India](/blog/meteorological-data-for-air-modelling-india/) cover the practical options, including the end of public NOAA ISD access discussed in [this update](/blog/noaa-isd-data-aermod-2025/).\n\n### Low Wind Speeds and Calms\n\nMany parts of India see long periods of very low wind on winter nights, and this is one of AERMOD's known weak points. Calm hours are not modelled in the usual way, and very low wind stable hours can give high predictions because the plume is barely diluted. The model includes options to adjust the friction velocity in stable conditions (ADJ_U* in AERMET), which became a regulatory option in later versions, and further low-wind options that remain non-default.\n\nThe practical point is that for a site with a high proportion of calm or near-calm hours, the 24-hour peak can be sensitive to these choices. Report the percentage of calm hours, state which options were used, and if the result is close to a limit, test the alternative. [Atmospheric stability classes explained](/blog/atmospheric-stability-classes-explained/) gives the background on why stable hours matter.\n\n### Background Concentrations\n\nFor NAAQS compliance, the increment is added to a baseline background, which in most Indian industrial areas is far larger than the increment from a single new stack. A 24-hour PM10 increment of 3 µg/m³ added to a background of 85 µg/m³ is a very different story from the same increment added to 45 µg/m³.\n\nThis means the uncertainty in the background often dominates the uncertainty in the final total. A one-season monitoring campaign, a monitor placed next to a road, or using the average rather than the 98th percentile can all shift the total more than any modelling choice. The [cumulative air quality impact assessment](/blog/cumulative-air-quality-impact-assessment-india/) guide covers how to combine background and increment defensibly.\n\n### Terrain and Land Use\n\nOn flat sites terrain adds little uncertainty, but near hills or elevated villages ignoring it can substantially understate concentrations where the plume meets high ground. Wrong surface characteristics (roughness, Bowen ratio, albedo) change AERMET's turbulence and mixing height estimates, and with them the size and location of peaks. See [terrain and land use data for AERMOD](/blog/terrain-land-use-data-aermod/) for the inputs involved.\n\n## Sensitivity Testing: How to Find Out What Matters\n\nSensitivity testing means changing one input at a time, rerunning the model, and seeing how much the key result moves. It is the most practical way to understand uncertainty in a specific study, and it costs little once a base case is set up.\n\n1. **Define the key outputs.** Maximum 24-hour and annual increments, and increments at named sensitive receptors, for each pollutant.\n2. **Fix and document a base case.** This is the case you will report.\n3. **Choose inputs that are both uncertain and influential.** Emission rate, exit velocity and temperature, meteorological year or station, surface characteristics, low wind options, terrain and background.\n4. **Use plausible alternatives, not arbitrary percentages.** For example, design emission against permitted limit, a second year of meteorology, or an alternative station.\n5. **Change one thing at a time** so each effect is attributable.\n6. **Tabulate the change** as a percentage of the base case, noting whether the maximum moved.\n7. **Interpret against the margin.** If every variation stays well below the standard, the conclusion is robust. If one plausible variation approaches it, that input needs better data or mitigation.\n\n![Tornado chart showing how much the predicted maximum 24-hour increment changes when each input is varied in an illustrative sensitivity test](/blog-images/aermod-guides/how-accurate-is-aermod-sensitivity-tornado.svg)\n\n## Worked Example: A Sensitivity Test for a Boiler Stack\n\nThe following example is **illustrative**. The numbers are illustrative, to show the method, and do not come from any real project.\n\nA proposed 40 TPH coal-fired boiler at a food processing unit in central India has a single stack. The CTE application needs to show that PM10 and SO2 increments, added to baseline, remain within NAAQS 2009. The base case uses:\n\n- PM10 emission rate based on the stack limit applicable to the boiler, converted to g/s using the design normal flue gas flow.\n- One year of surface meteorology from a station about 18 km away, processed through AERMET with ADJ_U* applied.\n- Flat terrain, rural dispersion, a nested Cartesian grid out to 10 km plus discrete receptors at six villages.\n- Baseline PM10 98th percentile of 72 µg/m³ from the monitoring campaign.\n\nThe base case predicts a maximum 24-hour PM10 increment of 4.0 µg/m³ about 1.2 km north-east of the stack. Total: 72 + 4.0 = 76.0 µg/m³ against the 24-hour standard of 100 µg/m³.\n\nThe sensitivity runs give the following (all **illustrative**):\n\n| Variation | Max 24-h PM10 increment (µg/m³) | Change from base | Location of max | Total with background (µg/m³) |\n|---|---|---|---|---|\n| Base case | 4.0 | n/a | 1.2 km NE | 76.0 |\n| Emission at measured typical rate (lower) | 2.6 | -35% | 1.2 km NE | 74.6 |\n| Exit velocity 20% lower (reduced plume rise) | 4.6 | +15% | 0.9 km NE | 76.6 |\n| Alternative meteorology station (35 km) | 5.8 | +45% | 1.6 km E | 77.8 |\n| ADJ_U* not applied | 5.0 | +25% | 0.8 km NE | 77.0 |\n| Surface roughness for cultivated land instead of mixed | 3.7 | -8% | 1.3 km NE | 75.7 |\n| Background at 98th percentile of a nearby CAAQMS (higher) | 4.0 | 0% | 1.2 km NE | 88.0 |\n\nSeveral lessons come out of this table:\n\n- **The model result is robust in this case.** Even the most pessimistic modelling variation (alternative meteorology, +45%) gives a total of 77.8 µg/m³, well below 100.\n- **Background matters most for the total.** Changing the background source moves the total by 12 µg/m³, far more than any modelling choice.\n- **Meteorology is the most influential modelling input.** The alternative station moves both the value and the location. That is worth a sentence in the report, and a justification of why the chosen station is more representative.\n- **The location of the peak shifts.** Report receptor-level results alongside the grid maximum.\n\nNow imagine a background of 94 µg/m³. The base case total becomes 98.0 µg/m³ and the alternative meteorology case 99.8 µg/m³, barely within the standard. The honest course is to say so, consider additional control such as better particulate control or a taller stack (see our [stack height calculation guide](/blog/stack-height-calculation-cpcb-guidelines/)), and agree the approach with the reviewing authority.\n\n## Comparing Model Predictions With Monitoring Near Operating Plants\n\nFor expansion projects, plants seeking consent to operate renewal, or cases where a regulator questions a result, you may have monitoring data near an operating plant. This is a chance to check the model, but it must be done carefully.\n\n### Why Direct Comparison Is Tricky\n\nA monitor measures every source, from traffic to regional haze; the model predicts only the sources you entered. A monitor reading of 85 µg/m³ PM10 and a model increment of 6 µg/m³ are not in conflict.\n\n### A Practical Approach\n\n1. **Choose a pollutant dominated by the plant.** SO2 near a coal-fired plant is better than PM10, which has many other sources.\n2. **Use upwind and downwind monitors.** Their same-day difference gives a rough estimate of the plant's contribution.\n3. **Model with actual emissions and concurrent meteorology.** Use CEMS or stack monitoring data and meteorology for the same period, ideally on-site.\n4. **Compare distributions, not hours.** Rank the estimated plant contribution and the predicted values at the monitor, draw a Q-Q plot and compute FB on the robust highest values.\n5. **Look for patterns.** Overprediction at every monitor may mean the emission rate is overstated; poor agreement at one monitor may mean the wind data misrepresents that part of the area.\n6. **Report the limitations:** short periods, non-plant sources and monitor siting.\n\nA good result here is not perfect agreement. A Q-Q plot where the top ranked values sit inside the factor-of-two lines, with a modest bias, is evidence the model set-up is reasonable for that site. Our guide to [air quality compliance reporting against NAAQS](/blog/air-quality-compliance-reporting-naaqs/) covers how monitoring statistics such as the 98th percentile should be calculated.\n\n## How to Communicate Uncertainty in an EIA\n\nIndian EIA reports rarely discuss model uncertainty beyond a boilerplate sentence. A short, specific discussion helps the Expert Appraisal Committee (EAC) or SEAC see that the conclusion is robust, and protects the proponent if monitoring later shows different values.\n\nThe EIA Notification 2006 and the Terms of Reference (ToR) issued for the project usually require prediction of impacts on ambient air quality using a suitable model, with the incremental GLCs superimposed on baseline. They do not prescribe a format for uncertainty, which leaves room for good practice.\n\n| Report section | What to include | Example wording (adapt to your study) |\n|---|---|---|\n| Input basis | Whether emissions are permitted limits, design values or measured | \"Emission rates correspond to the permitted stack limits at full load and therefore represent an upper bound.\" |\n| Meteorology | Station, period, data capture, calm percentage and representativeness | \"Calm hours were 14% of the period; ADJ_U* was applied as per current regulatory practice.\" |\n| Sensitivity | A short table of the key variations and the effect on the maximum | \"The maximum 24-hour increment varied from 2.6 to 5.8 µg/m³ across tested variations.\" |\n| Conclusion | Whether the compliance conclusion holds across the range | \"In all variations the total remains below the NAAQS 2009 24-hour PM10 standard.\" |\n\nSome practical rules:\n\n- **Avoid false precision.** Report increments to one decimal place at most, and do not report totals to two decimals.\n- **Separate increment and background.** Always show both, and the total.\n- **Name the conservative assumptions.** If emission limits, full load and continuous operation were assumed, say so.\n- **Do not claim site-specific accuracy without site-specific evidence.** You can say the model is extensively evaluated and is the US EPA preferred model for this application.\n\nFor more on what a complete report should contain, see [air dispersion modelling report contents](/blog/air-dispersion-modelling-report-contents/) and the [QA/QC checklist for air dispersion reports](/blog/qa-qc-air-dispersion-report/).\n\n## When AERMOD Is Not the Right Tool\n\nPart of using a model accurately is knowing its limits. AERMOD is a near-field model, generally applied up to about 50 km, and it assumes a single representative wind field each hour. Coastal sites with sea breezes, deep valleys and long-range questions may justify CALPUFF or a grid model. AERMOD also treats pollutants as inert, apart from NO to NO2 conversion options such as ARM2, OLM and PVMRM, so secondary PM2.5 and ozone need other methods. See our [AERMOD vs CALPUFF guide](/blog/aermod-vs-calpuff-or-air-dispersion-model-guide/) and [choosing an air dispersion model](/blog/choosing-an-air-dispersion-model/). Set-up errors such as unit mix-ups, wrong coordinates, missing downwash and coarse grids are covered in [common AERMOD mistakes](/blog/common-aermod-mistakes/), and each can change a result more than the model's inherent uncertainty.\n\n## How DSTechnoverse Helps\n\nWe set up AERMOD studies so that the uncertainty is understood rather than hidden: a documented emission basis, a justified meteorological dataset, receptor grids that catch the peak, and a short sensitivity table where the margin to the standard is tight. Our [AERMOD modelling service](/services/aermod-air-dispersion-modelling/) supports proponents and EIA consultancies, including white-label work. If you prefer to run cases yourself, [AERMOD Studio](/services/aermod-software/) runs the real US EPA AERMOD and AERMET in the browser, which makes repeat sensitivity runs quick. To discuss a project, [talk to our team](/contact/).\n\n## Frequently Asked Questions\n\n**Is AERMOD accurate enough for an Indian EIA?** Yes, for its intended purpose. AERMOD is the US EPA preferred model for near-field regulatory dispersion and has been evaluated against many field datasets. It reliably estimates the high end of the concentration distribution. In Indian studies, the accuracy of the final result depends far more on emission rates, meteorology and background data than on the model's physics, so input quality is where effort should go.\n\n**What does \"within a factor of two\" mean?** It means the predicted value lies between half and double the observed value. For regulatory models it is usually applied to the highest concentrations compared unpaired in time and space. A model meeting this test might predict a peak of 6 µg/m³ when the true peak is anywhere from 3 to 12 µg/m³, although well-configured AERMOD studies often agree more closely than that.\n\n**Why does my monitor show much higher values than the model?** Because the monitor measures all sources, including traffic, other industry, construction dust, biomass burning and regional pollution, while the model predicts only the sources you entered. The model increment should be added to a representative background before comparing with NAAQS. A direct comparison of increment against monitor reading will always suggest underprediction, even when the model is performing well.\n\n**What is the robust highest concentration?** The robust highest concentration (RHC) is a statistic used in US EPA model evaluation to estimate the peak concentration without being dominated by a single extreme value. It fits an exponential tail to the top ranked values, commonly the top 26, and calculates a stable estimate of the highest. Comparing predicted and observed RHC gives a fairer measure of peak performance than comparing single maxima.\n\n**How many meteorological years should I use?** US regulatory practice typically uses five years of representative off-site data or at least one year of on-site data. In India, one year or one season is common because of data availability. If you use a single period, a sensitivity run with a second year or an alternative station shows whether the conclusion depends on that choice.\n\n**Should I include a sensitivity analysis in my EIA report?** It is not a standard ToR requirement, but it is good practice, particularly when the total concentration is close to a NAAQS limit or the meteorological data is from a distant station. A short table showing the maximum increment under three to six plausible variations takes little space and gives the appraisal committee confidence that the conclusion is robust.\n\n**Can AERMOD handle the low wind conditions common in India?** Partly. Calm hours below the threshold are not modelled in the normal way, and very stable low wind hours can produce high predictions. AERMET's ADJ_U* option, now a regulatory option, improves stable low wind performance. Report the percentage of calm hours, state the options used, and test alternatives if the result is sensitive to these conditions.\n\n**Can monitoring data be used to validate my model?** For an operating plant, yes, with care. Choose a pollutant dominated by the plant, estimate the plant contribution using upwind and downwind monitors, model with actual emissions and concurrent meteorology, and compare ranked distributions using a Q-Q plot and fractional bias. For a new plant, validation is not possible, so sensitivity testing is the main tool.\n\n---\n\n**Need an AERMOD study or AERMOD software?** DSTechnoverse runs regulatory [AERMOD modelling studies](/services/aermod-air-dispersion-modelling/) for EC, CTE and expansion applications, for industries and as a white-label partner for EIA consultancies, and provides [AERMOD Studio](/services/aermod-software/), our AERMOD software, [personalised for your firm](/services/aermod-software-provider-india/). We are based in **Indore, Madhya Pradesh** and work across India. [Talk to our team](/contact/).","related":[{"slug":"air-dispersion-modelling-for-industries-india","title":"Air Dispersion Modelling for Industries in India: A Compliance Guide"},{"slug":"air-dispersion-modelling-thermal-power-plants-india","title":"Air Dispersion Modelling for Thermal Power Plants in India"},{"slug":"air-dispersion-modelling-services-india","title":"Air Dispersion Modelling Services in India: The Complete Guide"},{"slug":"air-quality-modelling-mining-projects-india","title":"Air Quality Modelling for Mining Projects in India"}]}