AERMET: processing raw weather into model-ready files

The previous page covered what the boundary-layer parameters mean. This page covers the machine that produces them: AERMET, the meteorological preprocessor whose use is mandatory for regulatory AERMOD — Appendix W directs that "AERMET shall be used to preprocess all meteorological data, be it observed or prognostic" (§8.4.2(a); quoted in full in Meteorological data & representativeness).

Four kinds of input data

AERMET processes four types of data: 1) hourly surface observations that are typically, but not exclusively, collected at airports by the National Weather Service (NWS) and/or the Federal Aviation Administration (FAA); 2) twice-daily upper air soundings collected by the NWS; 3) data collected from an on-site or site-specific measurement program; and 4) prognostic meteorological data processed through a processor such as the Mesoscale Model Interface (MMIF) (EPA, 2024; Ramboll, 2023).

Source · AERMET User's Guide (EPA-454/B-26-005), version 26135 §1.1 — captured verbatim in the source library.

A typical NWS-data application uses the first two together — a surface station plus an upper-air station — because the morning sounding is what the convective mixing height is grown from. Site-specific (onsite) and prognostic data get their own page in this Part.

The two stages

AERMET's two stages (UG §1.1) — four data types in, two AERMOD-ready files out

The first stage extracts the surface and upper air data from files in which the data are stored in specific archive formats. The quality of the surface, upper air, and site-specific or prognostic data is also assessed during Stage 1. The second stage reads the output from Stage 1, calculates the boundary layer parameters required by AERMOD, and generates two AERMOD-ready meteorological data files.

Source · AERMET UG (EPA-454/B-26-005) §1.1. Since version 21DRF the former three-stage design (extract/QA → merge → PBL) is two stages, runnable separately or in a single AERMET run; the merge stage is gone (§1.1).

Two side-doors feed Stage 2, and both matter for the quality of the result:

Hourly-averaged 1-minute ASOS winds from AERMINUTE can replace the standard archive's hourly winds for 2000 to present — the UG calls them "more representative than the surface wind data in the standard hourly archive formats" (§1.1). Why that is true is the AERMINUTE page's story.
Surface characteristics (roughness, Bowen ratio, albedo) "can be calculated from processors such as AERSURFACE (EPA, 2026b) for observed data and from the prognostic model when processing prognostic data" (§1.1) — the AERSURFACE page covers how.

Quality assessment is a review loop, not a filter

AERMET's QA "identifies occurrences of missing data, values that are outside a range of values, and inconsistencies between selected variables within an observation period" (§1.1.1) — but deliberately does not fix anything:

At present there are no provisions for AERMET to automatically replace missing data or correct "suspect" values. The user should review the QA messages and determine if the value(s) require modification or if they are acceptable.

Source · AERMET UG (EPA-454/B-26-005) §1.1.1.

And the UG draws a bright line about where any correction may happen: "Modifications should only be done on extracted data, and not on the archive or raw data file. The archived or raw data should never be altered but should be maintained as delivered" (§1.1.1). The raw record stays pristine; edits live in the extracted working copy, and re-QA follows every edit.

What comes out

The second stage … computes the boundary layer scaling parameters (e.g., surface friction velocity, mixing height, and Monin-Obukhov length), and produces two input files for AERMOD. The first file contains the computed boundary layer parameters, as well as the observed surface parameters (e.g., temperature, wind speed, and wind direction). The second file contains one or more levels (a profile) of winds, temperature, and the standard deviation of the fluctuating components of the wind if provided.

Source · AERMET UG (EPA-454/B-26-005) §1.1.2.

These are the surface (.sfc) and profile (.pfl) files AERMOD names on its ME pathway — the hand-off described on the previous page.

Where the data archives live now

The archive landscape shifted recently, and the 26135 UG documents the current state (§1.1.1):

Upper air: the FSL format is gone — "FSL data is no longer available beginning in autumn of 2024" — and the current archive is IGRA (the Integrated Global Radiosonde Archive at NOAA NCEI), supported since 21DRF.
Surface: alongside ISHD, AERMET reads GHCNh (Global Historical Climatology hourly) — the format "which has replaced the Integrated Surface Hourly Data (ISHD) format" at NCEI, per EPA's 2026-07-09 transmittal memo for the 26135 release.
Legacy formats (CD-144, SAMSON, HUSWO) remain readable for archival records; TD-3280 was dropped in 21DRF.

Version 26135 also adds upper-air substitution: AERMET can "read in a second upper air station and allow substitution of missing upper air data either from the primary upper air station and/or secondary station" (§1.4) — a direct answer to sounding gaps that previously punched holes in a met record.

The regulatory guardrails

Two Appendix W constraints complete the picture (both verified against the current eCFR):

Representativeness is judged per input. Addendum A's AERMOD entry requires that AERMET's inputs "possess an adequate degree of representativeness," judged "independently for each variable," with the surface characteristics reflecting the vicinity of the meteorological tower (or representative grid cell for prognostic data) — Addendum A, A.1 AERMOD, b.(2)(i).
The adjusted-u* option has a boundary. AERMET's ADJ_U* option — an alternative friction-velocity formulation for stable, low-wind hours added in version 16216 (Model Formulation §3.3.2) — requires alternative- model approval under §3.2 "for applications involving the use of site-specific meteorological data that includes turbulences parameters (i.e., sigma-theta and/or sigma-w)" — Addendum A, A.1 AERMOD, b.(2)(ii).

And the master guardrail from the DFAULT rule: regulatory AERMOD requires "meteorological data processed with the regulatory options in AERMET" (Addendum A, A.1 AERMOD, Regulatory Use — see Rule — Regulatory Default (DFAULT)).

In PlumeSmart

The met workspace runs this pipeline — current-format archives (ISHD/GHCNh surface, IGRA upper air), AERMINUTE where 1-minute data exist, AERSURFACE surface characteristics — with the regulatory options, and keeps the station pairing and processing record visible for review.