Data Sources

Data Sources

The global data layer that feeds CAOS_SEISMIC. Every source listed here is open, scriptable, and ingestible in an automated daily job. The page states, for each source: what it gives, how to access it, its license / attribution obligation, its update cadence, and its feasibility for daily inference.

Honest framing. The dominant signal in every published short-term forecasting result is the earthquake catalog itself (ETAS-class clustering). Geodesy, fault geometry, slab geometry, stress, and tides are upside, not foundation — they ship only if they show positive, significant pseudo-prospective information gain over a catalog-only baseline. This page therefore separates the catalog spine (load-bearing) from the complementary global enrichers (covariates), and treats tidal stress as a regularized covariate with an honestly small, regime-dependent effect size.

CAOS_SEISMIC is a global forecaster: it trains on global seismicity plus every complementary global covariate, and runs inference across many countries — high-seismicity (Chile, Japan, Indonesia, Mexico, Türkiye, California, New Zealand) and low-seismicity (United Kingdom, Germany, Australia, Brazil) — so that bias toward high-seismicity zones can be measured rather than assumed. The data layer is correspondingly global.

---

1. Architecture of the data layer

flowchart TD
    subgraph SPINE["Catalog spine (load-bearing)"]
        COMCAT["USGS ComCat (FDSN event)<br/>real-time global, daily"]
        ISCREV["ISC Bulletin REVIEWED + ISC-EHB<br/>relocated, ~24-month lag"]
        ISCGEM["ISC-GEM v12.1<br/>Mw-homogenized, 1904-2021"]
        GCMT["GCMT (.ndk)<br/>moment tensors, Mw>=5"]
        EMSC["EMSC SeismicPortal<br/>independent cross-check"]
    end
    subgraph ENRICH["Complementary global enrichers (covariates)"]
        SLAB2["Slab2<br/>subduction geometry"]
        FAULTS["GEM Active Faults<br/>fault distance / style"]
        PB2002["Bird PB2002<br/>plate boundaries"]
        GNSS["NGL GNSS / MIDAS<br/>strain-rate field"]
        WSM["World Stress Map<br/>stress orientation"]
        TIDES["Tidal stress (computed)<br/>pygtide + SPOTL/TPXO"]
    end
    SPINE --> HYG["Catalog hygiene<br/>(Mc, homogenize, decluster)"]
    ENRICH --> FEAT["Feature build<br/>(joined on the forecast grid)"]
    HYG --> FEAT
    FEAT --> MODEL["Conditional model<br/>(ETAS reference + neural challenger)"]

Loading

The spine is consumed every day; the enrichers are mostly static or slowly varying and are joined onto the forecast grid at build time. ObsPy's FDSN client is the unifying access layer for all catalog providers (see §3).

---

2. Catalog spine

The earthquake catalog is the foundation. Short-term forecasting skill scales with how low and stable the magnitude of completeness Mc is, so the global product is driven by ComCat with regional networks surfaced through it, and anchored on a long-term homogeneous reference for b-value and large-event statistics.

2.1 USGS ComCat — the spine (real-time global)

ANSS Comprehensive Earthquake Catalog: the single most important real-time, no-auth, daily-current global source. It is the spine of the daily inference loop.

Field	Value
What	Global event catalog (time, lat, lon, depth, magnitude, magType, IDs, products).
Access	FDSN event web service; libcomcat (Python) or ObsPy Client("USGS").
Base URL	https://earthquake.usgs.gov/fdsnws/event/1/
Formats	geojson (preferred for scripting), quakeml, csv, text, kml.
License	US Government work, public domain (USGS-authored content). Non-US networks surfaced through ComCat keep their own attribution.
Cadence	Real-time / continuously updated. Stability: very high.
Daily feasibility	Excellent. updatedafter returns only events updated since the last successful run — the natural incremental daily delta.

Operational notes.

• Hard cap: 20,000 events per request (exceeding it returns HTTP 400). Call count first (cheap) to decide whether to tile by time slice or magnitude band, then stitch and dedupe.
• magType is a first-class field — read and KEEP it. Silently mixing mb / Ms / Mw distorts the Gutenberg–Richter tail; the homogenization step (see Pipeline) depends on it.
• Be polite: serialize requests, prefer updatedafter deltas, and back off on HTTP 429/503. The public web app never hits upstream services directly — only the daily job pulls and persists.

# Count first (cheap) to decide whether to tile
curl "https://earthquake.usgs.gov/fdsnws/event/1/count?starttime=2010-01-01&endtime=2026-06-16&minmagnitude=2.5"
# Daily incremental: only events updated since the last successful run
curl "https://earthquake.usgs.gov/fdsnws/event/1/query?format=geojson&updatedafter=2026-06-15T00:00:00&minmagnitude=1"

from libcomcat.search import search
from libcomcat.dataframes import get_summary_data_frame
from datetime import datetime
events = search(starttime=datetime(2010, 1, 1), endtime=datetime(2026, 6, 16),
                minmagnitude=2.5)
df = get_summary_data_frame(events)   # id, time, lat, lon, depth, magnitude, magtype, ...

2.2 ISC Bulletin (REVIEWED) + ISC-EHB — clean retrospective bulletin

The International Seismological Centre is the definitive retrospective global bulletin: relocated and re-reviewed, the gold standard for completeness and homogenization. It lags real time by ~24 months, so it is a training-quality source, not a daily one.

Field	Value
What	Definitive relocated/reviewed global bulletin. The ISC-EHB subset gives the cleanest hypocenters.
Access	Catalogue web service http://www.isc.ac.uk/cgi-bin/web-db-run (`out_format=CATCSV
License	Open for research with attribution (check per-product terms).
Cadence	~24-month lag. Stability: very high.
Daily feasibility	Not a daily source. Use the REVIEWED DB for training quality; ISC throttles large pulls.

2.3 ISC-GEM v12.1 — long-term homogeneous anchor

The Global Instrumental Earthquake Catalogue: the catalog to use for long-term b-value, large-event recurrence, and as the Mw-homogenization anchor in the catalog hygiene step.

Field	Value
What	1904–2021, M ≥ 5.5 globally (continental events down to M 5.0), Mw-homogenized, relocated.
Version / DOI	v12.1 (2025-11-27). DOI 10.31905/d808b825.
Access	CSV (main + supplementary) + KMZ from https://www.isc.ac.uk/iscgem/ → download.php (also via GEM).
License	CC-BY-SA 3.0 (Unported) — © International Seismological Centre & GEM Foundation.
Cadence	Versioned (annual-ish), DOI'd. Stability: very high.
Daily feasibility	Periodic refresh, not daily.

Share-alike obligation. CC-BY-SA 3.0 share-alike means any redistributed derivative catalog must carry CC-BY-SA + attribution. Internal training is unaffected; if the product ships a processed ISC-GEM-derived catalog, the license and provenance are preserved. See §6 Attribution.

2.4 GCMT — moment tensors / mechanism enricher

Centroid moment tensors for essentially all M ≳ 5 global events since 1976: focal mechanism, Mw, nodal planes, P/T axes. The source of mechanism and stress features, and a Mw anchor alongside ISC-GEM.

Field	Value
What	Moment tensors (Mw, nodal planes, P/T axes) for ~all M ≳ 5 events since 1976.
Access	Download .ndk from https://www.globalcmt.org/CMTfiles.html (full file + monthly files since 2005 + Quick CMT). Parse natively with obspy.read_events(...) (obspy.io.ndk).
Format	NDK — ASCII, 5 lines × 80 chars per event.
License	Free for research with citation (Dziewonski et al. 1981; Ekström et al. 2012).
Cadence	Monthly batch + near-real-time Quick CMT. Stability: high.
Daily feasibility	Monthly refresh suffices; Quick CMT for near-real-time mechanism on large events.

2.5 EMSC SeismicPortal — independent cross-check

The European-Mediterranean Seismological Centre: best for low-latency real-time and an independent dedup cross-check against ComCat.

Field	Value
What	Independent real-time feed (Euro-Med + global) + felt reports; cross-check vs ComCat.
Access	FDSN-event (https://www.seismicportal.eu/fdsnws/event/1/query?..., json/xml/text, 20k cap); WebSocket push feed; ObsPy Client("EMSC").
License	Open for research with attribution.
Cadence	Real-time + WebSocket. Stability: high.
Daily feasibility	Excellent as a cross-check; call query?... directly (the bare /fdsnws/event/1/ index 404s).

2.6 Regional networks (surfaced through FDSN)

Regional networks dramatically beat the global catalog at small magnitudes (lower, more stable Mc). For a global product they are surfaced through the same FDSN layer and used as higher-resolution context where available; for any single country view they sharpen Mc. Examples with their attribution / redistribution status:

Region	Network	Access	License / flag
Chile	CSN (net C/C1)	EarthScope/IRIS FDSN or ISC; GNSS at gps.csn.uchile.cl	Public use with mandatory CSN attribution (since 2019-06)
California	SCEDC (CI), NCEDC (BK/NC)	Client("SCEDC")/Client("NCEDC"); AWS open-data s3://scedc-pds (--no-sign-request)	Open + attribution; AWS Open Data terms
New Zealand	GeoNet	https://service.geonet.org.nz/fdsnws/...; Client("GEONET")	CC-BY 3.0 NZ — attribute "Earth Sciences New Zealand"
Italy	INGV ISIDe (IV)	http://webservices.ingv.it/fdsnws/event/1/query; Client("INGV")	CC-BY (verify version on the dataset page)

Networks that are registration- and agreement-gated and not freely redistributable are used for internal training quality only, never shipped in the public product or web app.

---

3. ObsPy as the unifying access layer

obspy.clients.fdsn.Client hits every FDSN provider with one API. Built-in short names include USGS, ISC, EARTHSCOPE, GEONET, INGV, EMSC, SCEDC, NCEDC (IRIS/IRISDMC redirect to EARTHSCOPE). Operational rules:

• get_events() has no bulk analogue → loop time windows yourself and respect each provider's 20,000-event cap. Pattern: count → if > 20k, halve the window recursively → concatenate → dedupe by event id.
• Wrap every call in retry/backoff: HTTP 204 (no data), 400 (bad request), 413 (too large → "tile smaller"), 429/503 (slow down).

from obspy.clients.fdsn import Client
from obspy import UTCDateTime
cli = Client("USGS")                 # or "ISC", "EARTHSCOPE", "GEONET", "INGV", "EMSC", "SCEDC", "NCEDC"
cat = cli.get_events(starttime=UTCDateTime("2024-01-01"), endtime=UTCDateTime("2024-12-31"),
                     minmagnitude=2.5, orderby="time-asc")

---

4. Complementary global enrichers

These are geophysical-context covariates joined as static or slowly varying spatial fields onto the forecast grid. They are ranked by expected lift for a conditional short-term forecast (the ranking is inferred, not measured — each enricher's marginal information gain over a catalog-only baseline is quantified on held-out data before it is committed). None substitutes for the catalog.

4.1 Slab2 — subduction geometry (highest-value static covariate)

Depth-to-slab, dip, strike, thickness, and interface distance for every subduction zone. This is what separates interface, intraslab, and crustal seismicity — very different physics and rates — and is essential for the global product's tectonic-regime conditioning (see Technical Architecture).

Field	Value
What	Per-zone depth/strike/dip/thickness/uncertainty grids at 0.05° × 0.05°.
Access	USGS ScienceBase item 5aa1b00ee4b0b1c392e86467; code at github.com/usgs/slab2; NetCDF .grd.
License	USGS public domain.
Cadence	Static (versioned). Size: ~100s MB. Daily feasibility: loaded once.
Reference	Hayes et al. (2018), Science 362, 58–61. DOI 10.1126/science.aat4723.

4.2 GEM Global Active Faults + Bird PB2002 plate model

Tectonic-setting features: distance-to-nearest-active-fault, fault style, distance-to-plate-boundary, boundary type (subduction / transform / ridge), plate-pair relative velocity.

Source	What	Access	License	Reference
GEM Active Faults	Homogenized global active-fault database (GeoJSON/GPKG/KML/SHP)	github.com/GEMScienceTools/gem-global-active-faults	CC-BY-SA 4.0 (verify repo LICENSE)	GEM Science Tools
Bird PB2002	52 plates (14 large + 38 small); plate-boundary geometry + type	peterbird.name/publications/2003_pb2002/ (ASCII)	Open for research	Bird (2003), G-cubed 4(3), 1027. DOI 10.1029/2001GC000252

Both are static/versioned, small-to-moderate, loaded once.

4.3 NGL GNSS / MIDAS — strain-rate field

The Nevada Geodetic Laboratory processes >13,000 global GNSS stations with GipsyX. Interpolating station velocities yields a strain-rate field (dilatation, max shear, second invariant) that correlates with long-term seismicity rate — primarily a background covariate.

Field	Value
What	Daily E/N/U time series (.tenv3) + robust MIDAS velocities (midas.IGS14.txt).
Access	geodesy.unr.edu; wget/curl scriptable.
License	Open with attribution (NGL; Blewitt et al. 2016, MIDAS estimator, DOI 10.1002/2015JB012552).
Cadence	Daily–weekly. Size: GBs (per-station text). Daily feasibility: good (daily-updated, scriptable).

wget http://geodesy.unr.edu/velocities/midas.IGS14.txt
wget http://geodesy.unr.edu/gps_timeseries/tenv3/IGS14/<STATION>.tenv3

A ready-made alternative is the GEM/Kreemer GSRM global strain-rate model — a finished strain grid that is often easier than interpolating NGL stations directly.

4.4 World Stress Map / focal-mechanism stress

Crustal stress orientation and faulting regime. Derived from focal mechanisms (GCMT and regional moment tensors) via stress inversion, plus the World Stress Map compilation. Yields rake / faulting style, P/T-axis orientation, and Coulomb stress-transfer (ΔCFS) triggering features — physically motivated but second-tier effort. Open with attribution; small; event-driven.

4.5 Tidal stress (computed feature)

Tidal stress is computed, not downloaded — see the dedicated §5 Tidal triggering section below. It is a regularized covariate with an honestly small, regime-dependent effect.

4.6 Deferred / out of scope (v1)

Enricher	Why deferred
InSAR (COMET LiCSAR)	Heaviest enricher (TB rasters, atmospheric noise, frame-by-frame coverage); the deformation → short-term-probability link is weak and research-grade. Not in the daily loop; budget separate TB-scale storage out of git if added later.
Heat flow (World Heat Flow DB)	Sparse, static; at best a slow background covariate for crustal regions. Lowest priority.

---

5. Tidal triggering

Tides are a physically-motivated, regularized covariate with an honestly small effect — never a standalone predictor. This section states the physics, the effect sizes by regime, and the role of the feature in the model. It is the most over-claimed phenomenon in popular discourse and the most disciplined to handle in an honest forecaster.

5.1 The physics

Tidal stress on a fault has two distinct contributions that must be computed separately and summed:

1. Solid-Earth (body) tides — the Moon and Sun deform the whole solid Earth, producing a strain/stress field throughout the lithosphere. Dominant constituents: semidiurnal M2 (12.421 h), S2 (12.000 h); diurnal O1 (25.819 h), K1 (23.934 h); and the long-period fortnightly Mf (~13.66–14.77 d) and monthly Mm.
2. Ocean tidal loading (OTL) — the redistributed weight of ocean water flexes the seabed and adjacent crust. Near continental margins and on the ocean floor this can dominate the body tide. A 2 m tidal column imposes a vertical seafloor load of order $\rho_w g h \approx 1025 \times 9.81 \times 2 \approx 2.0\times10^{4}$ Pa ≈ 20 kPa, attenuating inland and with depth. This is why coastal/subduction thrusts show the strongest "tidal" signal — it is really ocean-loading stress, not body tide.

Modeling consequence. A correct tidal-stress series = body tide (tidal-potential catalog + Love numbers) + ocean loading (global ocean-tide model convolved with elastic Green's functions). Doing only the body tide underestimates the signal at coastal/subduction sites by a large factor.

Resolve to the fault — tidal Coulomb failure stress. Form the full tidal stress tensor $\sigma_{ij}(t)$ at the hypocenter (body + OTL), then resolve it onto the fault plane with unit normal $\hat{n}$ and slip direction $\hat{s}$:

$$\Delta\tau(t) = \hat{s}\cdot\big(\sigma(t),\hat{n}\big), \qquad \Delta\sigma_n(t) = \hat{n}\cdot\big(\sigma(t),\hat{n}\big)$$

$$\Delta\mathrm{CFS}(t) = \Delta\tau(t) + \mu_f,\Delta\sigma_n(t)$$

with $\mu_f \approx 0.4$ the (apparent) friction coefficient (a poroelastic variant replaces $\mu_f$ with an effective $\mu' = \mu_f(1 - B)$, Skempton coefficient $B$, for fluid-rich subduction interfaces).

5.2 Why the effect is small — and where it is larger

The decisive numbers:

Quantity	Order of magnitude
Tidal stress on faults	~0.1–10 kPa (body ~1 kPa; OTL up to ~10–20 kPa at strong-tide coast/seafloor)
Earthquake static stress drop	~0.1–10 MPa (typically 1–10 MPa)
Ratio tidal/stress-drop	~$10^{-3}$ to $10^{-4}$

A tide therefore cannot cause an earthquake; at most it can advance or retard the timing of a rupture already within ~$10^{-3}$ of failure — a small minority at any instant. The nucleation- duration argument (Beeler & Lockner 2003) is the decisive one: if nucleation takes longer than a tidal period, the oscillating stress averages out and no phase preference survives. Lab + rate-state extrapolation put crustal nucleation on a major fault at ~1 year ≫ daily tide, so daily tides should not strongly correlate — only the fortnightly Mf band, or settings with very short nucleation, should show a clean correlation. The signal is largest where (a) ocean loading is strong (margins, ridges, seafloor), (b) faults are shallow (less lithostatic clamping), and (c) the dominant tidal component aligns with the failure mode (thrust ↔ vertical ocean load; ridge normal faults ↔ ridge-axis tension).

5.3 Effect sizes by regime (canonical literature)

Study	Setting	Effect size
Tanaka, Ohtake & Sato (2002), JGR — DOI 10.1029/2001JB001577	9,350 global M ≥ 5.5	No correlation when lumped; reverse (thrust) faults correlate with tidal shear stress
Cochran, Vidale & Tanaka (2004), Science — DOI 10.1126/science.1103961	Shallow ocean-loaded thrusts	Rate varies by factor ~3; ~75% occur during encouraging stress at tides > 2 m; best at $\mu = 0.4$; chance < 1 in 10,000. California: only ~1–2%.
Métivier et al. (2009), EPSL — DOI 10.1016/j.epsl.2008.12.024	442,412 global events	~0.5–1.0% rate excess at favorable phase, ~99% confidence; larger for smaller/shallower events
Ide, Yabe & Tanaka (2016), Nat. Geosci. — DOI 10.1038/ngeo2796	Tidal stress vs event size	b-value decreases as tidal shear amplitude rises; 2004 Sumatra, 2010 Maule, 2011 Tohoku near peak. Contested (Hough 2018 found no lunar-phase effect for 204 M ≥ 8).
Scholz, Tan & Albino (2019), Nat. Comms — DOI 10.1038/s41467-019-10605-2	Mid-ocean ridges	Phase inverts via axial magma-chamber response; clean rate-state $R/r=\exp(\Delta\mathrm{CFS}/A\sigma)$ with no triggering threshold

The much stronger, least-disputed channel is not regular earthquakes at all — it is nonvolcanic tremor and slow-slip events, which respond to the same ~kPa stresses far more strongly (exponential tremor rate, fortnightly Mf modulation used as an in-situ stressmeter; Rubinstein et al. 2008 DOI 10.1126/science.1150558; Houston 2015 DOI 10.1038/ngeo2419; van der Elst et al. 2016 DOI 10.1073/pnas.1524316113). Where tremor/SSE catalogs exist (Cascadia, Nankai, Mexico, parts of Chile/Hikurangi), the slow-earthquake channel is kept separate from the fast-earthquake channel in the model.

5.4 The rate-state functional form

The physically correct shape for the covariate is exponential, not linear (Dieterich 1994; Scholz et al. 2019):

$$\frac{R}{r} = \exp!\left(\frac{\Delta\mathrm{CFS}}{A\sigma}\right)$$

so the tidal channel enters the conditional intensity as a regularized multiplier:

$$\lambda(t \mid \mathcal{H}) = \lambda_0(t\mid\mathcal{H})\cdot \exp!\Big(\beta,\tfrac{\Delta\mathrm{CFS}(t)}{A\sigma}\Big)$$

where $\lambda_0$ is the base conditional model and $\beta \in [0,1]$ is a learned, regularized coupling the data is allowed to drive to ~0. For the size channel (Ide), the forecast b-value may depend on tidal shear amplitude, $b(t) = b_0 - \kappa,A_{\tau,\text{tide}}(t)$, with $\kappa$ also regularized toward 0.

5.5 Computing the feature

Tool	Component
pygtide (github.com/hydrogeoscience/pygtide, wraps ETERNA PREDICT 3.4)	Body-tide strain → stress via the crustal stiffness tensor $\sigma_{ij}=\lambda,\epsilon_{kk}\delta_{ij}+2G,\epsilon_{ij}$
SPOTL (Agnew; nloadf/hartid)	Ocean tidal loading via mass-loading Green's functions
TPXO (+ GOT/FES alternatives)	Global ocean-tide model used as the OTL load

Engineered features: $\Delta\mathrm{CFS}(t_0)$ at the forecast instant; its rate $\dot{S}(t_0)$; phase as $\sin\theta,\cos\theta$ (circular); semidiurnal envelope; and the fortnightly Mf (~14.7 d) envelope.

Validation gate (mandatory before the feature touches any public number). Reproduce a published tidal time series for a known site (e.g. Cascadia / Axial Seamount), confirm sign conventions (tension- vs compression-positive) and the Green's-function reference frame (CE vs CM) match across pygtide and SPOTL, then require positive prospective CSEP information gain (with-vs-without, declustered) — never claim improvement from in-sample Schuster p-values. License: pygtide / SPOTL are open tools; TPXO has its own academic-use terms (recorded). At coastal/subduction margins, ocean loading dominates — skipping it is the single biggest modeling error.

5.6 Honest expectation

• Most regions: $\beta \to \approx 0$, negligible skill gain, possibly a tiny calibration improvement at the favorable phase — reported honestly, never buried.
• Shallow ocean-loaded thrust / ridge regions: measurable but small (~few %) improvement, useful mostly for calibration, not sharpness.
• Regions with tremor/SSE: the covariate can meaningfully sharpen the slow-slip forecast, which indirectly informs megathrust stress state.

---

6. Attribution and licensing obligations

The product credits page must display every required attribution; these are tracked so the credits notice is not forgotten.

Source	Obligation
USGS / ANSS ComCat	Public domain (US Government work); credit USGS/ANSS.
ISC-GEM v12.1	CC-BY-SA 3.0 — share-alike: a redistributed derived catalog keeps CC-BY-SA + provenance.
ISC Bulletin / ISC-EHB	Open for research with attribution.
GCMT	Free for research with citation (Dziewonski 1981; Ekström 2012).
EMSC SeismicPortal	Open for research with attribution.
CSN (Chile)	Public use with mandatory CSN attribution (since 2019-06).
GeoNet (NZ)	CC-BY 3.0 NZ — attribute "Earth Sciences New Zealand."
INGV ISIDe	CC-BY (verify exact version on the dataset page).
Slab2 / USGS	Public domain; credit USGS.
GEM Active Faults	CC-BY-SA 4.0 (verify repo LICENSE).
Bird PB2002	Open for research with citation.
NGL GNSS / MIDAS	Open with attribution (NGL; Blewitt et al. 2016).
TPXO ocean-tide model	Academic-use terms (recorded).

These are not redistribution blockers — but the public app must display them. Tokens, restricted-network credentials, and any agreement-gated raw files live only in a private secrets vault and are never in the public repo or web app.

---

7. Update cadence and daily-inference feasibility (summary)

Source	Cadence	Daily-inference role
USGS ComCat	Real-time	Spine — daily updatedafter delta
EMSC SeismicPortal	Real-time + WS	Daily cross-check
GCMT (Quick CMT)	Monthly + near-RT	Mechanism on large events
NGL GNSS / MIDAS	Daily–weekly	Strain-rate covariate (slow)
ISC REVIEWED / ISC-EHB	~24-month lag	Training quality (not daily)
ISC-GEM v12.1	Versioned (annual-ish)	Mw anchor / b-value (periodic refresh)
Slab2, GEM faults, Bird PB2002	Static	Loaded once; tectonic-regime conditioning
Tidal stress (computed)	On demand	Regularized covariate (computed per inference)

Bottom line: ComCat + EMSC give a fully scriptable, no-auth, real-time daily spine; the homogeneous anchors and the geophysical enrichers are periodic or static and are refreshed on a slower cadence. The whole layer is reproducible from manifests (see Pipeline) and feeds the system described in Technical Architecture.

---

References

• Bird, P. (2003). An updated digital model of plate boundaries (PB2002). G-cubed 4(3), 1027. DOI 10.1029/2001GC000252.
• Blewitt, G., Hammond, W. C., & Kreemer, C. (2016). MIDAS robust trend estimator for GPS station velocities. JGR Solid Earth 121. DOI 10.1002/2015JB012552.
• Cochran, E. S., Vidale, J. E., & Tanaka, S. (2004). Earth tides can trigger shallow thrust fault earthquakes. Science 306(5699), 1164–1166. DOI 10.1126/science.1103961.
• Di Giacomo, D., Engdahl, E. R., Storchak, D. A., et al. ISC-GEM Global Instrumental Earthquake Catalogue v12.1. DOI 10.31905/d808b825.
• Dieterich, J. (1994). A constitutive law for the rate of earthquake production and its application to earthquake clustering. JGR 99(B2), 2601–2618. DOI 10.1029/93JB02581.
• Beeler, N. M., & Lockner, D. A. (2003). Why earthquakes correlate weakly with the solid Earth tides. JGR Solid Earth 108(B8), 2391. DOI 10.1029/2001JB001518.
• Ekström, G., Nettles, M., & Dziewoński, A. M. (2012). The global CMT project 2004–2010. PEPI 200–201, 1–9. DOI 10.1016/j.pepi.2012.04.002.
• Hayes, G. P., et al. (2018). Slab2, a comprehensive subduction zone geometry model. Science 362, 58–61. DOI 10.1126/science.aat4723.
• Houston, H. (2015). Low friction and fault weakening revealed by rising sensitivity of tremor to tidal stress. Nat. Geosci. 8, 409–415. DOI 10.1038/ngeo2419.
• Ide, S., Yabe, S., & Tanaka, Y. (2016). Earthquake potential revealed by tidal influence on earthquake size–frequency statistics. Nat. Geosci. 9, 834–837. DOI 10.1038/ngeo2796.
• Métivier, L., et al. (2009). Evidence of earthquake triggering by the solid earth tides. EPSL 278, 370–375. DOI 10.1016/j.epsl.2008.12.024.
• Rubinstein, J. L., et al. (2008). Tidal modulation of nonvolcanic tremor. Science 319, 186–189. DOI 10.1126/science.1150558.
• Scholz, C. H., Tan, Y. J., & Albino, F. (2019). The mechanism of tidal triggering of earthquakes at mid-ocean ridges. Nat. Comms 10, 2526. DOI 10.1038/s41467-019-10605-2.
• Tanaka, S., Ohtake, M., & Sato, H. (2002). Evidence for tidal triggering of earthquakes. JGR Solid Earth 107(B10), 2211. DOI 10.1029/2001JB001577.
• van der Elst, N. J., et al. (2016). Fortnightly modulation of San Andreas tremor and low-frequency earthquakes. PNAS 113(31), 8601–8606. DOI 10.1073/pnas.1524316113.

Service docs: USGS FDSN-event earthquake.usgs.gov/fdsnws/event/1/ · ISC web services isc.ac.uk/iscbulletin/search/webservices/catalogue/ · ObsPy FDSN docs.obspy.org/packages/obspy.clients.fdsn.html · GeoNet FDSN geonet.org.nz/data/access/FDSN · EMSC SeismicPortal seismicportal.eu/webservices.html · Slab2 earthquake.usgs.gov/slab2/ · GEM faults github.com/GEMScienceTools/gem-global-active-faults.

---

See also: Pipeline — how this data is cleaned, homogenized, and turned into features · Technical Architecture — how the global field, regime tiling, and daily job consume it.