Preliminary GeoClaw results for 11 March 2011 Honshu event

Most of the links to simulation results still work.

We need better bathymetry and topography near Japan to better resolve effects near the coast. 500m bathymetry provided by Barry Eakins is posted on the bathymetry wiki page. We are stil working on processing this data into the form we can use, but this is still insufficient resolution for modeling inundation or comparing against near-shore tide gauges.

Any help in obtaining better data would be appreciated.

These are preliminary results obtained with GeoClaw, not yet fully validated.

We will try to make codes available that are used for these simulations in hopes they will be useful to others, but this is on-going work and may have flaws so use at your own risk. If you obtain results that you would like to contribute, or find errors in codes, please contact Randy LeVeque.

For links to documentation and some preprints of recent papers on the algorithms uses, please see the GeoClaw website.

The codes used will appear in the tsunami-benchmarks Subversion repository. You can obtain the current state of many of these codes via

    svn checkout http://kingkong.amath.washington.edu/svn/tsunami-benchmarks

We are collecting more data, to be posted on the wiki and/or in honshu

Contributions of other tools or data most welcome!

Run times are reported below for some computations, and refer to time on a single processor of either a desktop or MacBook. The more recent ones with -O3 optimization, some of the earlier runs were without optimization and should have run more quickly.

Posted 20 March 2011:

• DART 21418 (closest to Japan), with 1-minute resolution of UCSB-3 fault model and 30-second resolution of tsunami, using 5 levels of AMR. (Run time: 14 minutes)
  Simulation ... DART 21418 ... code in repository
  Note: This computation and the ones posted earlier are all assume an instantaneous desplacement during the earthquake. The time history of the fault model is not yet incorporated and may cause a change in these results. This is currently under development.

• First attempt at Crescent City tide gauge: Simulation ... Tide gauge comparison ... GeoClaw code used for this simulation. (Run time: 2.5 hours)
  This tide gauge is in the harbor. With GeoClaw the wave appears to arrive about 10 minutes too early but the first wave has nearly the correct amplitude and period. We fail to capture the largest 4th wave, it's not clear yet how this is generated. We are using 1-minute resolution on level 6 of AMR but only very close to the harbor. Moreover the topography does not appear to fit together well from two different resolution topo files used in this computation. We are continuing to work on this problem.
• Comparison to DART buoys south of Japan: Simulation ... DART comparisons ... GeoClaw code used for this simulation. (Run time: 1.5 hours)
• Note that DART 51425 is mis-labeled as 52425 in the above comparison, and shows a similar displacement in time as the Crescent City tide gauge. A map of all the DART buoys and real-time data are available at http://www.ndbc.noaa.gov/dart.shtml
  

Posted 18 March 2011:

• Chen Ji at UCSB pointed out a significant difference between the plots for DART 21413 posted on 16 March with 2-minute and 4-minute resolution and those posted 14 March with 1-minute resolution. Running the comparions just for this location does show quite a bit of difference between 2-minute and 4-minute also, which appears to be due to differences in the resolution used near the source region. This is under investigation but the convergence claimed on 16 March may not be so clear.

  Some new plots (resolution refers to finest grids on AMR level 4):
     1-minute vs. 2-minute shortly after quake ... 1-minute vs. 2-minute at DART 21413
     4-minute vs. 2-minute shortly after quake ... 4-minute vs. 2-minute at DART 21413

Posted 16 March 2011:

• Simulation of initial wave approaching Fukushima nuclear power plant, where power was lost approximately 55 minutes (0.92 hours) after the quake. This test was motivated by a post on the TBB, and appears to show that it was the first wave that inundated this region. Computed with 5 levels of AMR on 1-minute bathymetry, and dx=dy= 4-arcsecond on finest level. (Run time = 15 minutes).

• Comparison of results at many more DART buoys: Simulation with 4 levels of AMR and 2 minute resolution on finest level ... DART Comparisons. (Run time = 8 hours) This is a repeat of the computation of many DART buoys posted on 14 March, but with twice the resolution and approximately 8 times longer run time (2 cubed, since grids at all levels are refined in x,y, and t).

  The results look virtually the same. This indicates that the solution is essentially converged. The fact that the simulated tide gauges do not agree exactly with the actual DART data is due to a combination of various factors, such as:

  • The earthquake source (Version 3 of the UCSB model) is not exactly correct,
  • The shallow water equations are not a perfect model for the wave motion,
  • The 4-minute resolution etopo ocean bathymetry is may not be fine enough, though in view of the resolution of the computational grid and the vastly longer wavelength of the tsunami, this is presumably a more minor effect than the first two.
  • The criteria used for adaptive refinement captures the leading wave well but uses much less resolution for the trailing waves and reflections. The behavior at the DART buoys behind the initial wave therefore does not have as much structure as one would see on a uniform grid of the finest resolution, and different results would be seen on a uniform grid.

Posted 14 March 2011:

• DART buoy 21413 comparsion redone with Preliminary Version 3 of the UCSB source (link below)
  Simulation using UCSB source, with 4 levels of refinement and 1-min resolution on the finest grids. ... DART comparison

• Comparison of results at many more DART buoys: Simulation with 4 levels of AMR and 4 minute resolution on finest level ... DART Comparisons. (Run time = 52 minutes)

Posted 13 March 2011:

Our first attempt at modeling a tide gauge on the coast of Japan gives curious results. The gauge is at 43.28N 145.57E on the eastern tip of Hokkaido. The Tide gauge record shows the first three waves arising at times that are about double those shown by the GeoClaw gauge. This may be due to lack of resolution of the bathymetry in this region. This Google Earth image shows that the region is very shallow and there is an island in the way of the tsunami that does not show up on the 1-minute bathymetry used. The current GeoClaw results may be totally spurious.

• Simulation using UCSB source, with 5 levels of refinement and 1-min resolution on the finest grids. ...
• GeoClaw code used for these simulations.

Posted 12 March 2011:

A preliminary comparison of the tsunami generated by by the USGS and UCSB sources listed below.

• Near Japan -- USGS source (on left), UCSB source (on right)

• Sendai Airport -- using USGS source and ETOPO 1-minute bathymetry.
• Sendai Airport -- using UCSB source and ETOPO 1-minute bathymetry.
• GeoClaw code used for these simulations.

DART buoy 21413 comparsion

• Simulation using UCSB source, with 4 levels of refinement and 1-min resolution on the finest grids. ... DART comparison
• GeoClaw code used for these simulations.

Posted 11 March 2011:

Calculated with USGS source:

• Japan to Chile
• Japan to Hawaii
• Better resolution near Japan

• USGS source parameters from http://earthquake.usgs.gov/earthquakes/eqinthenews/2011/usc0001xgp/finite_fault.php ... Subfault Format

• UCSB source parameters from http://www.geol.ucsb.edu/faculty/ji/big_earthquakes/2011/03/0311/Honshu.html ... Subfault Format

• UCSB source parameters from http://www.geol.ucsb.edu/faculty/ji/big_earthquakes/2011/03/0311/Honshu_main.html now contains more recent versions of this source.

• In either case, parameters were processed with the Okada model to obtain seafloor deformation, using subfaults.py (Revision 1). So far the varying rupture times and rise times have been ignored, and a static displacement is specified at time t = 1 s.