Hurricane / Tropical-Cyclone Study Arc — Data Sources
Status: active arc · Opened: 2026-06-09 · Owner: Mike + Claude (content development)
This is a measured-reality arc. The bottom line: any data we ingest has to be something that happened in reality. That means direct readings from real instruments, the physical record of events that actually occurred. We ingest no model output of any kind, from us or anyone else: no forecasts, no projections, no reanalysis, no objective-analysis grids, nothing a computer predicts or infers. We are not in the business of divining the future. We study the physical fingerprint a storm leaves across every instrument that measures it — river gauges, tide gauges, barometers, buoys, lightning-detection networks, seismometers, radiosondes, in-situ ocean floats, direct satellite radiometers — across the full lifecycle: forming → intensifying → peak → weakening → post-storm.
The bright-line test for any source: is this a measurement of something that physically happened, or is it a computer's estimate of what will, might, or "would have" happened? The first is in; the second is out. A modeled river-discharge forecast is out; a stream gauge reading the actual river is in. A reanalysis pressure grid is out; a barometer on a buoy is in.
Working principle for this arc: we fill data gaps to make a paper work; we do not drop a paper because a feed isn't wired up yet. Every "needed but not yet ingested" source below is a candidate fetcher, not a reason to abandon the idea. (This does not relax the bright line: a gap gets filled with a real-measurement source, never with model output.)
Taxonomy reference: each paper is tagged by the two axes (data velocity × window type) from
docs/paper-taxonomy.md.
Are all tropical-cyclone seasons the same? (the seed fact)
No. Each basin runs its own calendar, and the hemispheres are ~6 months out of phase, so there is almost always an active season somewhere:
| Basin | Season | Peak | Notes |
|---|---|---|---|
| North Atlantic | Jun 1 – Nov 30 | ~Sep 10 | The "hurricane season" everyone means |
| Eastern Pacific | May 15 – Nov 30 | ~Aug–Sep | Starts two weeks earlier than Atlantic |
| Western Pacific (typhoons) | no official bounds; year-round | Jul–Oct | Busiest basin on Earth (~⅓ of global TCs) |
| North Indian | bimodal: Apr–Jun and Oct–Dec | May, Nov | Summer monsoon shears the mid-season dead patch |
| Southern Hemisphere (S Indian / Australian / S Pacific) | Nov – Apr | Feb–Mar | ~6 months out of phase with the NH |
| South Atlantic | effectively none | — | One famous storm: Catarina, Mar 2004 |
The closest thing to a true global lull is roughly May. This fact is the basis of Paper 1.
The papers in this arc
P1 — Global season-phase monitor (open / standing-monitor · slow velocity)
Question: Where on Earth is tropical-cyclone activity at any moment? Quantify the hemispheric 6-month offset, the North Indian double peak, and whether a truly storm-free day ever happens. Status: fully sourceable today (IBTrACS is live back to 1842). Second entry in the open/standing-monitor quadrant after the Starlink pilot.
P2 — Freshwater pulse: the landfall water budget (closed window per storm · slow velocity)
Question: What is the lag from landfall to river crest, and the total streamflow anomaly a storm injects into a basin? US-coast story (no global river feed). Status: sourceable today on real stream gauges (USGS streamflow, live). The modeled flood discharge (Open-Meteo) is excluded under the bright line; we use the gauges reading the actual river, not a discharge model.
P3 — Barometric & surge autopsy (closed window per storm · medium velocity)
Question: The raw pressure-minimum trace through a storm's life, and the surge footprint it stamps on the coast. Tide gauges and offshore buoys are the ground truth. Status: surge side is live on real tide gauges (NOAA tides). The pressure and wave side must come from real barometers and wave buoys, not Open-Meteo's weather/wave model (excluded): that promotes NDBC buoys (in-situ sea-level pressure + measured wave height) and NOAA CO-OPS stations from nice-to-have up to a P3 backbone. IGRA radiosondes give real upper-air pressure for context. With NDBC added, P3 runs entirely on instruments.
P4 — Lightning as a rapid-intensification tell (closed-or-open · fast velocity)
Question: Does a burst of eyewall lightning precede intensification? An open question in the literature, attackable with raw strike data. Status: live going forward (Blitzortung, but only ~7 weeks of history as of 2026-06-09). Add GOES GLM for satellite-side eyewall mapping and pre-April retrospective.
P5 — Hearing the hurricane (microseism) (closed-or-open · fast velocity)
Question: A storm churning the ocean generates secondary microseisms (~0.1–0.35 Hz) that show up on seismometers hundreds of miles inland. Can we watch a hurricane's "hum" rise and fall on a seismograph, and correlate it with wave height and storm distance? Status: needs a new seismic feed. We have earthquake catalogs, not the continuous ground-motion trace this requires. This is the one paper gated on a new fetcher — and the one we most want.
Cross-cutting fuel & confound layers
Used as supporting layers across P1–P5, all sourced as real measurements: sea-surface temperature from in-situ buoys and Argo floats and direct satellite radiometers (the fuel), subsurface ocean heat from Argo profiles (rapid-intensification fuel), the Saharan Air Layer from direct satellite imagery (Atlantic suppression), and scatterometer ocean winds from measured radar backscatter (raw surface wind under the storm). Note: the common gridded SST products (e.g. OISST) are objective-analysis blends, partly model-assisted, so we prefer the in-situ and direct satellite readings under the bright line.
Sources already live in TerraPulse
Real measurements — eligible (no work needed)
These are instruments recording events that physically happened. All clear the bright line.
| Source (slug) | Instrument / what it measures | Coverage (as of 2026-06-09) | Used by |
|---|---|---|---|
ibtracs_hurricanes |
Best-track record of storms that occurred (analyst-fixed positions + intensity from real observations; not a forecast) | 1842 → present, 815k recs | P1, P4, P5 |
usgs_water |
Stream gauges reading actual river stage / streamflow | Sep 2025 → present, 2.7M obs | P2 |
noaa_tides |
Tide gauges reading actual coastal water level → surge | Sep 2025 → present, 365k obs | P3 |
blitzortung_lightning |
Ground network detecting actual lightning strikes | Apr 20 2026 → present, 21M strikes | P4 |
igra_soundings |
Radiosondes measuring actual upper-air pressure / wind / structure | 2024 → present | P3 (context) |
nws_alerts |
Official warnings issued (a real event, but not a measurement) | 2021 → present | context only, not analysis input |
Excluded under the bright line — model / reanalysis output (live, but not eligible)
These are ingested in TerraPulse and fine for the platform generally, but they are computer estimates, not measurements, so they are out for this arc. Each has a real-instrument replacement listed in the gap-fill tiers.
| Source (slug) | Why excluded | Real-measurement replacement |
|---|---|---|
open_meteo (/v1/forecast) |
Weather model output (forecast / reanalysis fields) | NDBC buoys + CO-OPS + IGRA (real pressure/wind) |
open_meteo_flood (/v1/flood) |
GloFAS hydrological model discharge | usgs_water stream gauges (already live) |
open_meteo_marine (/v1/marine) |
Wave model output | NDBC buoys (measured significant wave height) |
era5_cloud |
ERA5 reanalysis (already inactive) | direct satellite radiometer / in-situ |
Sources to add (gap-fill, by priority)
URLs are canonical, free, and no-auth unless noted. Each row is a candidate fetcher.
Tier 1 — unlocks a gated paper
| Need | Source | URL | Notes |
|---|---|---|---|
| Seismic noise PSD (microseism band) | IRIS MUSTANG noise-psd / noise-pdf metrics |
https://service.iris.edu/mustang/ | Cheap path — pre-computed power spectral density per station/day; no waveform crunching. The 0.1–0.35 Hz band is the secondary microseism. Unlocks P5. |
| Raw seismic waveforms (BHZ vertical) | IRIS/FDSN dataselect web service |
https://service.iris.edu/fdsnws/dataselect/1/ | If we want to compute the PSD ourselves at higher cadence. Station metadata: https://service.iris.edu/fdsnws/station/1/ |
| Station discovery across networks | FDSN federated catalog | https://service.iris.edu/irisws/fedcatalog/1/ | Find inland broadband stations near US coasts for P5. |
Tier 2 — strengthens P3/P4 and the live-season story
| Need | Source | URL | Notes |
|---|---|---|---|
| In-situ ocean pressure/wind/wave (storm core) — now a P3 backbone | NOAA NDBC buoys (realtime2) | https://www.ndbc.noaa.gov/data/realtime2/ | Buoys literally inside the storm; direct measurement of sea-level pressure + wave height. Required for P3 now that the Open-Meteo pressure/wave model is excluded. Station list: https://www.ndbc.noaa.gov/to_station.shtml |
| Satellite lightning (GLM) | NOAA GOES-16/18 GLM on AWS Open Data | https://registry.opendata.aws/noaa-goes/ | Direct satellite detection of actual strikes; maps eyewall vs rainband lightning; covers the pre-April-2026 gap Blitzortung can't. |
| Live current-storm fixes | NHC CurrentStorms feed | https://www.nhc.noaa.gov/CurrentStorms.json | Atlantic + E Pacific. Ingest the observed current position/intensity only — NOT the forecast cone/track (forecast products are excluded). GIS: https://www.nhc.noaa.gov/gis/ |
| W Pacific / S Hemisphere current fixes | Joint Typhoon Warning Center (JTWC) | https://www.metoc.navy.mil/jtwc/jtwc.html | Fills the basins NHC doesn't cover. Observed fix only, not the forecast track. Products: https://www.metoc.navy.mil/jtwc/products/ |
| Finer coastal surge / water level | NOAA CO-OPS API | https://api.tidesandcurrents.noaa.gov/api/prod/ | Real tide gauges; higher-resolution companion to noaa_tides. |
Tier 3 — fuel & confound layers (cross-cutting)
| Need | Source | URL | Notes |
|---|---|---|---|
| Sea-surface temperature, in-situ (the fuel) | NDBC buoy + Argo SST (direct readings) | https://www.ndbc.noaa.gov/data/realtime2/ · https://data-argo.ifremer.fr/ | Preferred SST source. Real thermometers in the water. |
| Sea-surface temperature, direct satellite | NOAA/NESDIS L2 SST swaths (radiometer retrieval) | https://www.star.nesdis.noaa.gov/socd/sst/ | Direct measurement of emitted IR; eligible. Avoid the gridded OISST blend (https://www.ncei.noaa.gov/products/optimum-interpolation-sst) — it is an objective-analysis, model-assisted product, excluded under the bright line. |
| Subsurface ocean heat profiles | Argo float network | https://data-argo.ifremer.fr/ | In-situ temperature/salinity vs depth ahead of a storm track. Real floats. |
| Scatterometer ocean surface wind | NOAA/NESDIS ocean winds (ASCAT) | https://manati.star.nesdis.noaa.gov/ | Surface wind retrieved from measured radar backscatter; eligible. KNMI mirror: https://scatterometer.knmi.nl/ |
| Saharan Air Layer (Atlantic suppression) | CIMSS tropical / SAL split-window | https://tropic.ssec.wisc.edu/real-time/sal/ | Direct satellite imagery; ingest as a qualitative confound layer. (Coral Reef Watch's gridded heat-stress product is a derived analysis, so it is excluded; use Argo for subsurface heat instead.) |
Notes & honest limits
- Measured reality only: no model, forecast, projection, reanalysis, or objective-analysis output enters this arc, from us or anyone else. This excluded the three live Open-Meteo feeds (forecast / flood / marine) and ERA5. The cost is real: we trade convenient model grids for the work of wiring up real instruments (NDBC, CO-OPS, Argo, direct-satellite SST). The payoff is that every number in these papers is something that physically happened.
- NDBC is now load-bearing for P3: with the Open-Meteo pressure/wave model excluded, P3's pressure-minimum trace and wave height depend on getting NDBC buoys (and CO-OPS) wired up. This is a fill-the-gap task, not a reason to drop P3.
- Best-track is a record, not a forecast: IBTrACS and the NHC/JTWC observed fixes are analyst-determined from real observations, so they are eligible. Their forecast products (cones, predicted tracks) are not, and we do not ingest them.
- Lightning history is shallow: Blitzortung starts 2026-04-20. Any retrospective on earlier 2026 storms needs GLM (Tier 2).
- River/surge signals are US-only: USGS water, NOAA tides, NDBC, CO-OPS all cover US waters. P2 and P3 are US-coast stories unless we add foreign hydrology feeds.
- IBTrACS lags for the live storm: best-track is provisional for the current system; pair with NHC/JTWC (Tier 2) for true real-time intensity.
- Microseism is the one hard gate: P5 cannot start until the IRIS/FDSN feed (Tier 1) is wired. Everything else in this arc is doable with what's live or a light Tier-2 add.