[sgjholt]

Three bugs of one class, found by a live smoke fetch against IRIS rather than by the suite: the config promised a selection the download did not honour.

Each is a property ObsPy's MassDownloader supplies for free and the thin wrapper had to build by hand — sanitize, CircularDomain, channel_priorities.

1. The radius never reached the waveforms

max_radius_km was converted to degrees and passed to get_stations. The waveform call repeated the config's wildcards:

stream = client.get_waveforms(
    network=config.stations.network,
    station=config.stations.station,   # "*"
    ...
)

FDSN dataselect takes no geographic parameters, so this is not a missing argument — a wildcard waveform request is unbounded however the stations were selected.

On a Magna fetch with max_radius_km = 50, that returned three BH channels from UU.SRU, 225 km out. SRU is absent from the archived inventory.xml, which was radius-filtered, so the artefact held waveforms with no metadata beside them. min_radius_km had the identical hole.

Fix: the waveform request is built from the inventory the station query returned — one bulk line per channel, epochs collapsed, -- for a blank location code. The request is the metadata, so a trace without a station entry cannot be asked for.

2. A degree is not 111.195 km

The conversion was max_radius_km / 111.195. A degree of arc is not a fixed distance; measured on WGS84 across latitude and azimuth:

	km per degree
equator	110.574
poles	111.691
spread	1.01%

So one constant moves the boundary by −0.44% to +0.56% — ±0.3 km on a 50 km radius, ±2 km on Magna's 400 km.

Fix: the station query is sent at the loose end of that bracket, deliberately wide so the server cannot clip what the exact cut would keep, and the cut is made locally against the true epicentral distance from gps2dist_azimuth. max_radius_km = 50 now means 50 km wherever the event is. The two bracketing constants are asserted against the geodetics rather than trusted, so tightening them back toward 111.195 fails.

Verified down to sub-kilometre arrays: at a 500 m radius the cut resolves a 2 m difference in station position, and the widening is 3.1 m.

Also fixed here: min_radius_km was only sent alongside max_radius_km, so on its own it did nothing. A non-positive radius and an inverted annulus are now refused rather than silently selecting nothing — the failure mode of a metres-for-kilometres typo on a local array.

3. Every instrument at every station

channel = "HH*,BH*,HN*,EN*" returns all of them wherever they are installed — 507 channels for Magna's 88 stations. Since the request is built from that answer, all 507 are downloaded, and a station carrying a broadband beside an accelerometer contributes the same ground motion twice: two entries in a SpectrumSet, weighted as two stations by anything that averages over them.

Fix: channel_priorities and location_priorities on [stations], first-match-wins per station:

channel = "HH*,BH*,HN*,EN*"
channel_priorities = ["HH[ZNE]", "BH[ZNE]", "HN[ZNE]", "EN[ZNE]"]

channel says what the data centre may offer; the priorities say which of what it offers to keep. Applied in ObsPy's order — channels within a location, then locations across the station — so a list carried over from Restrictions selects the same instrument. A station matching no pattern is dropped, and the written inventory.xml is pruned to match.

With no list set the fetch warns and names the stations; ["*"] keeps everything and says so deliberately. datasets/magna_2020.toml stays unranked on purpose — the near-source broadbands are the ones that clip, and the accelerometer beside them is why the paper covers that distance range — and now records why.

The manifest gains channels_available and channels_requested beside channels, so what the query offered, what was asked for and what arrived are three separate numbers.

Tests

26 new tests. Those for §1 fail against the previous implementation; the fake client no longer offers get_waveforms at all, so a return to the wildcard path cannot pass.

Verification

pytest -m "not dataset and not notebook"    1207 passed, 15 skipped
pytest --without-optional-extras            1188 passed, 18 skipped
pytest tests/test_acquire.py                60 passed
ruff check / ruff format --check            clean
mypy                                        no issues in 65 source files
sphinx-build                                clean (bar blocked intersphinx here)

Not covered: a live re-fetch confirming the radius and the priorities hold on the wire. This sandbox has no FDSN egress, so that check is the reporter's to repeat — the channel count should drop and inventory.xml should account for every waveform file.

🤖 Generated with Claude Code

[read-the-docs-community]

Documentation build overview

📚 SpecMod | 🛠️ Build #34622737 | 📁 Comparing 2bd4fdc against latest (a7ad8bd)

  🔍 Preview build  

6 files changed · ± 6 modified

± Modified

• api.html
• genindex.html
• releasing-data.html
• roadmap.html
• tutorial/specmod-tutorial.html
• _modules/specmod/acquire.html