Site-specific towers & prognostic (WRF/MMIF) meteorology

National Weather Service (NWS) airport data is the workhorse, but Appendix W recognizes two other ways to feed AERMET: measurements from an on-site (site-specific) program, and prognostic model output processed through MMIF. This page covers when each applies and what the Guideline on Air Quality Models (Appendix W) demands of it. (Record lengths and the preference order are covered in Meteorological data & representativeness: 5 years NWS, ≥1 year site-specific — preferred when available — or ≥3 years prognostic, per §8.4.2(e).)

Site-specific data: preferred, but held to a standard

A tower at the site measures the atmosphere your plume actually disperses in — which is why §8.4.2(e) prefers it. The price is a real monitoring program. Appendix W's recommendations for site-specific data begin by pointing to a dedicated U.S. Environmental Protection Agency (EPA) guidance document — Meteorological Monitoring Guidance for Regulatory Modeling Applications (EPA-454/R-99-005) — and then set the floor:

As a minimum, site-specific measurements of ambient air temperature, transport wind speed and direction, and the variables necessary to estimate atmospheric dispersion should be available in meteorological datasets to be used in modeling. Care should be taken to ensure that meteorological instruments are located to provide an adequately representative characterization of pollutant transport between sources and receptors of interest.

Source · 40 CFR 51 App. W §8.4.4.2(a).

Three details from the same recommendations worth knowing before designing a program:

• Temperature is measured at standard shelter height (2 m), and solar or net radiation with a properly sited pyranometer/net radiometer (§8.4.4.2 i–ii).
• Cloud cover has a substitute. "AERMET may employ the Bulk Richardson scheme, which requires measurements of temperature difference, in lieu of cloud cover or insolation data" — with reviewing-authority consultation before using it (§8.4.4.2 iii). This is how a tower with ΔT sensors can drive the energy balance no human observer is there to provide.
• Wind should represent plume height. "Characterization of the wind profile up through the layer in which the plume disperses is desirable. This is especially important in complex terrain and/or complex wind situations" (§8.4.4.2 iv) — the per-variable representativeness idea from §8.4.2(b) made concrete.

In AERMET terms, these data enter the ONSITE pathway as user-described ASCII — there is no standard archive format for tower data, so the user defines the record layout and AERMET QA's it like everything else (AERMET UG §1.1.1).

Prognostic data: when no station will do

Appendix W frames prognostic meteorology as the answer to a specific problem:

For some modeling applications, there may not be a representative NWS or comparable meteorological station available (e.g., complex terrain), and it may be cost prohibitive or infeasible to collect adequately representative site-specific data. For these cases, it may be appropriate to use prognostic meteorological data, if deemed adequately representative, in a regulatory modeling application.

Source · 40 CFR 51 App. W §8.4.5.1(a).

What a prognostic model actually is

A prognostic model is the same kind of physics simulation that produces weather forecasts. It divides the atmosphere over a region into a three-dimensional grid and computes, cell by cell and hour by hour, how wind, temperature, and moisture evolve — reconstructing the weather everywhere on the grid, including places that have never had a weather station. For dispersion modeling it is run over past years rather than the future, with real observations blended in to keep the simulation anchored to what actually happened. EPA's MMIF guidance says it plainly:

In retrospective simulations (i.e., modeling past events), the blending of observed data with computed fields yields results that are bound by ground truth.

Source · EPA, Guidance on the Use of the Mesoscale Model Interface Program (MMIF) for AERMOD Applications (EPA-454/B-23-006), §2.

The two models the regulation names are generations of this same tool. WRF — the Weather Research and Forecasting model — is "the most used by EPA and the modeling community … supported across a broad community and provides state-of-the-science parameterizations of the atmosphere" (MMIF guidance §2). MM5 — the Fifth Generation Penn State/NCAR Mesoscale Model — is the earlier model of the same kind; the regulation still names it because archives of MM5 output remain usable. So when Appendix W says "MM5 or WRF data," read: the gridded output of a weather-physics model, older or current generation.

What MMIF does, and why it exists

A weather model's output is three-dimensional gridded fields in scientific data formats — nothing AERMET can read. The Mesoscale Model Interface (MMIF) is the translator EPA provides: it extracts the model output at your chosen grid cell — Appendix W notes MMIF "can process data for input to AERMET or AERMOD for a single grid cell or multiple grid cells" (§8.4.5.1(b)) — and rewrites it as AERMET-ready inputs. In effect, the grid cell plays the role a meteorological tower would: an hourly surface record plus wind and temperature profiles aloft.

From there the path is prescribed, and it runs through AERMET like every other data source: "when using MMIF to process prognostic data for regulatory applications, the data should be processed to generate AERMET inputs and the data subsequently processed through AERMET for input to AERMOD" (§8.4.5.1(b)); the MMIF guidance itself opens with the same instruction. Alternative processing methods need reviewing-authority approval. Inside AERMET, prognostic data uses the PROG pathway and "the same processing as ONSITE data" (AERMET UG §1.1 footnote); surface characteristics come from the prognostic model's own grid cell rather than AERSURFACE (App. W §8.4.2(b); AERSURFACE UG §1.1).

Prognostic met is not a shortcut, though — it carries its own burden of proof:

a. Prognostic model evaluation. Appropriate effort by the applicant should be devoted to the process of evaluating the prognostic meteorological data. The modeling data should be compared to NWS observational data or other comparable data in an effort to show that the data are adequately replicating the observed meteorological conditions of the time periods modeled. An operational evaluation of the modeling data for all model years (i.e., statistical, graphical) should be completed.

Source · 40 CFR 51 App. W §8.4.5.2(a) — use is "contingent upon the concurrence with the appropriate reviewing authority" that data quality has been demonstrated.

And the grid cell is a modeling decision: it "should be adequately spatially representative of the analysis domain. In most cases, this may be the grid cell containing the emission source of interest" (§8.4.5.2(b)).

Choosing between the three, in practice

The structure of §8.4 gives the decision order directly: use an adequately representative NWS station (5 years) unless representativeness fails; a site-specific program (≥1 year) is preferred whenever its data exist; and prognostic data (≥3 years, MMIF→AERMET, with an operational evaluation and reviewing-authority concurrence) is the documented route when neither observation source can represent the site. Whatever the source, §8.4.2(a)'s rule holds: all of it goes through AERMET.

In PlumeSmart

The platform's met workspace supports all three routes — NWS station pairing, uploaded on-site tower data on the ONSITE pathway, and prognostic (MMIF) data where observational coverage fails — and the record-length rule for the chosen route is checked against the years you select.