AERMET Tuning Guide

AERMET is the meteorological preprocessor for AERMOD. Bad met data silently produces bad dispersion results — this guide covers the choices and quality checks pyaermod surfaces to help you get them right.

Input data sources supported

pyaermod exposes four upstream met-data channels through pyaermod.met_ingest:

Source	Use for	Module
ISD / ISD-Lite (NOAA)	hourly surface observations at airport / METAR stations	ISDFetcher
ASOS 1-minute DSI-6405	sub-hourly wind data for AERMINUTE preprocessing	parse_asos_1min_file + aggregate_1min_to_hourly
IGRA v2	upper-air radiosonde soundings	IGRAFetcher + parse_igra_v2
MMIF	prognostic (WRF / MM5 / RAMS) fields in AERMOD-ready format	MMIFConfig

Cache directories are supported on both network fetchers so repeated runs for the same station / year don't redownload.

Stage 1, 2, 3 at a glance

AERMET runs in three stages:

1. Stage 1 ingests raw surface + upper-air data, applies basic range / consistency QA, and writes intermediate data files.
2. Stage 2 merges surface, upper-air, and (optionally) 1-minute ASOS data into a single hourly file.
3. Stage 3 computes boundary-layer parameters (Monin-Obukhov length, friction velocity, convective/mechanical mixing heights) and writes the final .SFC + .PFL files that AERMOD consumes.

pyaermod.aermet provides AERMETStage1, AERMETStage2, and AERMETStage3 dataclasses plus write_aermet_runfile() to emit each stage's input deck.

Quality assurance

Every AERMET run should be followed by a met_qaqc pass on the resulting .SFC output:

from pyaermod import read_surface_file, run_all_qaqc

records = read_surface_file("stn.sfc")["records"]
report = run_all_qaqc(records)
print(report.summary())
if report.n_errors:
    print(report.dump(limit=10))

run_all_qaqc runs five checks:

Check	What it catches
check_missing_data	>10% missing in any primary field; long (≥24 h) gaps
check_extremes	physically implausible values (T, wind, mixing height, u*, L)
check_stability_consistency	CBL regime (L<0) with zic=0, or SBL (L>0) with zic>0
check_low_wind_bias	>25% of hours at/below 0.5 m/s — often an anemometer artifact
check_profile_monotonic	non-decreasing pressure in a radiosonde ascent

Any finding at "error" severity means AERMOD will either fail or produce statistics you can't defend. "warning" findings are worth reviewing before the run.

Low-wind options

AERMOD's LOWWIND family adjusts how the model handles stable, calm hours. Appendix W (2017) currently allows LOWWIND3 when documented.

pyaermod.regulatory.EPA_APPENDIX_W_2017.check(project) will warn if low_wind_option is outside the allowed set.

ADJ_U* and surface characteristics

Stage 3 is where the friction-velocity adjustment (ADJ_U*) is enabled in AERMET's run file. AERMOD couples with this through profile inputs; pyaermod does not auto-enable it — set AERMETStage3.adj_ustar = True explicitly and include it in your modeling protocol.

Common AERMET failures

Symptom	Likely cause	Check / fix
AERMOD crashes with "missing hour" errors	Stage 2 couldn't produce a full annual record	find_missing_runs(records, "wind_speed_ms")
Very low predicted impacts	calm-bias inflating	check_low_wind_bias(records)
Predicted impacts ignore elevated plumes	profile file missing upper-air data	read_profile_file("stn.pfl") and inspect level count
Station far from project site	surface obs unrepresentative	document distance + terrain similarity; consider MMIF

Recommended workflow

from pyaermod import (
    ISDFetcher, ISDStationId, IGRAFetcher,
    AERMETStage1, AERMETStage2, AERMETStage3,
    read_surface_file, run_all_qaqc,
)

# 1. Fetch raw obs (cached)
surface = ISDFetcher(cache_dir="cache/isd").read_hourly(
    ISDStationId("723010", "13880"), 2020,
)
upper = IGRAFetcher(cache_dir="cache/igra").read_soundings("USM00072469")

# 2. Write AERMET run decks for stages 1-3 (see aermet.write_aermet_runfile)

# 3. Run aermet (external binary) for each stage

# 4. QA the Stage 3 output before feeding AERMOD
records = read_surface_file("stn.sfc")["records"]
report = run_all_qaqc(records)
if report.n_errors:
    raise SystemExit(report.dump())