About this service, coverage and limits
Researchers submit a tsunami source (one fault, or a set of subfaults) and receive a map and gridded data of the maximum water level offshore. The source is converted to initial seafloor displacement with the Okada (1985) elastic half-space solution, and a fast surrogate (neural-operator) model trained on numerical tsunami simulations predicts the maximum water level on the regional grid in seconds to minutes instead of hours.
The map colours the maximum water-surface elevation over 3 hours (not run-up) with a continuous scale that adapts to each event (default range: 1 % of its maximum up to the maximum, Linear or Log, manual minimum and maximum, metres or centimetres, location of the maximum). Optionally a full solver animation adds frames of the surface elevation every 5 minutes of model time, computed by the numerical solver the model was trained on, with a comparison of its maximum against the fast model.
Covered regions
20 subduction-zone regions, each with its own model trained on simulated megathrust ruptures inside an 8 x 8 degree box. 20 are available now.
| Region | Longitude | Latitude | Training Mw | Training events | Model version | Test error, offshore 10-50 m: median |ln(model/solver)| | Distance guard: warn / refuse (km) |
|---|---|---|---|---|---|---|---|
| Kamchatka | 156.0 to 164.0 | 47.0 to 55.0 | 7.6 to 9.1 | 1300 | fno-kamchatka-cd9eee88 |
0.123 (factor 1.13); 71 % of tiles within 20 % | 13 / 79 |
| Alaska (Aleutian, central) | 199.0 to 207.0 | 50.0 to 58.0 | 7.6 to 9.1 | 1300 | fno-alaska-f57f7da8 |
0.159 (factor 1.17); 61 % of tiles within 20 % | 11 / 67 |
| Mexico | 256.0 to 264.0 | 11.0 to 19.0 | 7.6 to 9.1 | 1300 | fno-mexico-af67626a |
0.134 (factor 1.14); 69 % of tiles within 20 % | 15 / 91 |
| Peru | 278.5 to 286.5 | -17.5 to -9.5 | 7.6 to 9.1 | 1300 | fno-peru-25d40cea |
0.126 (factor 1.13); 72 % of tiles within 20 % | 14 / 81 |
| Philippine Trench | 124.0 to 132.0 | 7.0 to 15.0 | 7.6 to 9.0 | 1300 | fno-philippines-b660d46f |
0.116 (factor 1.12); 74 % of tiles within 20 % | 12 / 72 |
| Hikurangi (New Zealand) | 174.0 to 182.0 | -44.0 to -36.0 | 7.6 to 9.0 | 1300 | fno-hikurangi-181e75f2 |
0.16 (factor 1.17); 61 % of tiles within 20 % | 14 / 83 |
| Cascadia | 230.0 to 238.0 | 41.0 to 49.0 | 7.6 to 9.1 | 1300 | fno-cascadia-4fc051ba |
0.146 (factor 1.16); 65 % of tiles within 20 % | 15 / 87 |
| Central-south Chile | 281.5 to 289.5 | -39.5 to -31.5 | 7.6 to 9.1 | 1300 | fno-chile-018d9d13 |
0.12 (factor 1.13); 73 % of tiles within 20 % | 12 / 75 |
| Sumatra (Sunda, north) | 91.5 to 99.5 | -2.0 to 6.0 | 7.6 to 9.1 | 1300 | fno-sumatra-3dab34bf |
0.128 (factor 1.14); 70 % of tiles within 20 % | 15 / 89 |
| Java | 105.0 to 113.0 | -14.0 to -6.0 | 7.6 to 9.1 | 1300 | fno-java-04ee6692 |
0.12 (factor 1.13); 72 % of tiles within 20 % | 14 / 82 |
| Colombia-Ecuador | 276.0 to 284.0 | -3.0 to 5.0 | 7.6 to 9.1 | 1300 | fno-colombia_ecuador-43769db7 |
0.093 (factor 1.1); 82 % of tiles within 20 % | 11 / 67 |
| Northern Chile | 282.0 to 290.0 | -25.5 to -17.5 | 7.6 to 9.1 | 1300 | fno-chile_north-08e20788 |
0.137 (factor 1.15); 67 % of tiles within 20 % | 14 / 82 |
| Ryukyu | 124.0 to 132.0 | 22.0 to 30.0 | 7.6 to 9.0 | 1300 | fno-ryukyu-9b96e159 |
0.165 (factor 1.18); 60 % of tiles within 20 % | 14 / 82 |
| Manila Trench | 116.0 to 124.0 | 14.0 to 22.0 | 7.6 to 9.0 | 1300 | fno-manila-cd2c0e89 |
0.133 (factor 1.14); 68 % of tiles within 20 % | 13 / 75 |
| Tonga | 182.0 to 190.0 | -26.0 to -18.0 | 7.6 to 9.1 | 1300 | fno-tonga-dcf4374f |
0.144 (factor 1.16); 66 % of tiles within 20 % | 12 / 69 |
| Andaman | 89.0 to 97.0 | 7.0 to 15.0 | 7.6 to 9.1 | 1300 | fno-andaman-13b321f2 |
0.14 (factor 1.15); 66 % of tiles within 20 % | 14 / 83 |
| Makran | 58.0 to 66.0 | 21.0 to 29.0 | 7.6 to 9.0 | 1300 | fno-makran-98d88080 |
0.145 (factor 1.16); 64 % of tiles within 20 % | 14 / 85 |
| Lesser Antilles / Puerto Rico | 295.0 to 303.0 | 13.0 to 21.0 | 7.6 to 9.0 | 1300 | fno-lesser_antilles-f112429a |
0.142 (factor 1.15); 66 % of tiles within 20 % | 16 / 97 |
| Tohoku (Japan Trench) | 138.5 to 146.5 | 34.0 to 42.0 | 7.6 to 9.1 | 1400 | fno-tohoku-075ac668 |
0.121 (factor 1.13); 72 % of tiles within 20 % | 15 / 89 |
| Nankai | 130.5 to 138.5 | 30.0 to 38.0 | 7.6 to 9.1 | 1400 | fno-nankai-4cc5c71e |
0.143 (factor 1.15); 65 % of tiles within 20 % | 16 / 98 |
Test error: the surrogate is compared with the numerical solver it was trained on, for simulated ruptures held out from training (200 per region), using the maximum water level in 0.2 degree tiles of the 10-50 m depth band; the table gives the median of the absolute natural-log ratio, and "factor" is its exponential. It is not a comparison with observed tsunamis. Each result page and README lists the model version and the full test summary of the region used. Only the offshore maximum water level is returned.
- Strike-slip, normal and oblique sources: the training sources are megathrust ruptures (rake close to 90 degrees).
- Crustal and other non-megathrust sources, including outer-rise events, splay faults, landslides and volcanic sources.
- Very deep sources (centroid below about 60 km) and sources far from the training ruptures of the region.
- Run-up, inundation, current speed and time series (arrival time, wave period): only the offshore maximum water level over 3 hours is returned.
- Locations outside the region boxes, and regions without a trained model.
- Slip heterogeneity beyond what you supply: see the note below.
Slip heterogeneity
A rupture with uniform slip on one rectangle (the default single-fault source) gives offshore maxima at the low end of the training events of the same Mw, which had heterogeneous slip (for a uniform-slip Tohoku Mw 9.0 rectangle the model gave 11.8 m, versus a median of 14.8 m for training events of similar Mw). Results from a uniform-slip rectangle therefore tend to underestimate heterogeneous ruptures of the same magnitude. The service does not offer a stochastic-slip option yet.
How to use
- Register with an academic e-mail address. Registration is automatic: you receive your personal access code on the page straight after registering (save it). Unusual or duplicate registrations are reviewed instead. See the terms and fair use.
- Log in with your e-mail and access code. On the map above, click a region, paint slip on the fault cells (or upload a CSV, or enter fault parameters), and press "Check and run". The classic forms stay available under Advanced.
- Review the routing and input checks, confirm any warnings, and submit. The job is queued and processed by a compute worker; the maximum offshore water level is drawn over the same map and can be downloaded.
Limits and input checks
- 20 events per user per 24 hours; 2 queued or running jobs at a time; fair use applies (terms).
- CSV source: up to 2000 subfaults and 1024 KB. Source extent up to 1500 km; Mw 5.0 to 9.7.
- Results are deleted after 30 days. Solver animations (1 queued or running per user, 3 per user per 24 hours, at most 15 in the queue, 25 minutes of computing time at most) are kept 7 days; they take a few minutes and run only when the compute machine is not busy.
- Sources outside all covered regions are refused; sources near an edge, spanning regions, or outside the training magnitude range are accepted only after you confirm the warning.
- Rake: refused outside 60 to 120 degrees; warning outside 75 to 105.
- Dip: refused outside 3 to 60 degrees; warning outside 5 to 45.
- Centroid depth: refused above 80 km; warning above 60 km or below 3 km.
- Distance of the source centroid to the nearest training source of the region: warning above the 99th percentile of the training nearest-neighbour distances (table above), refused above 6.0 times that value.
Disclaimers
- Trial service. Values are estimates from a machine-learning surrogate of a numerical solver, not outputs of a full hydrodynamic simulation, and they may be wrong for sources unlike the training ruptures.
- Bathymetry is the 30 arcsec grid used in training; local features smaller than a few kilometres are not resolved.
- The model domain is a box of 8 x 8 degrees. Along its four edges an absorbing strip of about 56 km (0.5 degree) damps the waves; results there are not reliable.
- The service is not an operational warning product. Do not use it for warning, evacuation or any life-safety decision; use official warning-centre products.
- The accuracy figures describe agreement with a solver on simulated events only.
Citation
[citation placeholder: Authors (Year), Title, Journal/Zenodo DOI]
Contact
[contact placeholder: name, email]