src/examples/tsunami.c

    The 2004 Indian Ocean tsunami

    The 2004 Indian Ocean tsunami was caused by a large-scale fault rupture (> 1000 km) at the Indian–Australian and Eurasian–Andaman plate boundaries. This example uses the fault model of Grilli et al, 2007 as initial conditions for a Saint-Venant solution of the subsequent tsunami. A similar setup is discussed in Popinet, 2011.

    Solver setup

    The following headers specify that we use spherical coordinates and the Saint-Venant solver together with (dynamic) terrain reconstruction and the Okada fault model.

    #include "spherical.h"
#include "saint-venant.h"
#include "terrain.h"
#include "okada.h"

    We then define a few useful macros and constants.

    int maxlevel = 10;
#define MINLEVEL 5
double ETAE  = 1e-2; // error on free surface elevation (1 cm)
double HMAXE = 5e-2; // error on maximum free surface elevation (5 cm)

    The maximum number of levels to use can be set as an argument to the program.

    int main (int argc, char * argv[])
{
  if (argc > 1)
    maxlevel = atoi(argv[1]);

    Here we setup the domain geometry. We choose to use metre as length unit, so we set the radius of the Earth (required for the spherical coordinates) in metres. The x and y coordinates are longitude and latitude in degrees, so we set the size of the box L0 and the coordinates of the lower-left corner (X0,Y0) in degrees.

      Radius = 6371220. [1];
  // the domain is 54 degrees squared
  size (54.);
  // centered on 94,8 longitude,latitude
  origin (94. - L0/2., 8. - L0/2.);

    G is the acceleration of gravity required by the Saint-Venant solver. This is the only dimensional parameter. We rescale it so that time is in minutes.

      // acceleration of gravity in m/min^2
  G = 9.81*sq(60.);

    When using a tree (i.e. adaptive) discretisation, we want to start with the coarsest grid, otherwise we directly refine to the maximum level. Note that 1 << n is C for 2^n.

    #if TREE
  // 32^2 grid points to start with
  init_grid (1 << MINLEVEL);
#else // Cartesian
  // 1024^2 grid points
  init_grid (1 << maxlevel);
#endif

    We then call the run() method of the Saint-Venant solver to perform the integration.

      run();
}

    We declare and allocate another scalar field which will be used to store the maximum wave elevation reached over time.

    scalar hmax[];

    Boundary conditions

    We set the normal velocity component on the left, right and bottom boundaries to a “radiation condition” with a reference sealevel of zero. The top boundary is always “dry” in this example so can be left alone. Note that the sign is important and needs to reflect the orientation of the boundary.

    u.n[left]   = - radiation(0);
u.n[right]  = + radiation(0);
u.n[bottom] = - radiation(0);

    Adaptation

    Here we define an auxilliary function which we will use several times in what follows. Again we have two #if…#else branches selecting whether the simulation is being run on an (adaptive) tree or a (static) Cartesian grid.

    We want to adapt according to two criteria: an estimate of the error on the free surface position – to track the wave in time – and an estimate of the error on the maximum wave height hmax – to make sure that the final maximum wave height field is properly resolved.

    We first define a temporary field (in the automatic variable η) which we set to h+z_b but only for “wet” cells. If we used h+z_b everywhere (i.e. the default \eta provided by the Saint-Venant solver) we would also refine the dry topography, which is not useful.

    int adapt() {
#if TREE
  scalar eta[];
  foreach()
    eta[] = h[] > dry ? h[] + zb[] : 0;

    We can now use wavelet adaptation on the list of scalars {η,hmax} with thresholds {ETAE,HMAXE}. The compiler is not clever enough yet and needs to be told explicitly that this is a list of doubles, hence the (double[]) type casting.

    The function then returns the number of cells refined.

      astats s = adapt_wavelet ({eta, hmax}, (double[]){ETAE,HMAXE},
			    maxlevel, MINLEVEL);
  fprintf (stderr, "# refined %d cells, coarsened %d cells\n", s.nf, s.nc);
  return s.nf;
#else // Cartesian
  return 0;
#endif
}

    Initial conditions

    We first specify the terrain database to use to reconstruct the topography z_b. This KDT database needs to be built beforehand. See the xyz2kdt manual for explanations on how to do this.

    We then consider two cases, either we restart from an existing snapshot or we start from scratch.

    The next line tells the Saint-Venant solver to conserve water surface elevation rather than volume when adapting the mesh. This is important for tsunamis since most of the domain will be close to “lake-at-rest” balance.

    event init (i = 0)
{
  terrain (zb, "~/terrain/etopo2", NULL);

  if (restore (file = "dump"))
    conserve_elevation();
  else {
    conserve_elevation();

    The initial still water surface is at z=0 so that the water depth h is…

        foreach()
      h[] = max(0., - zb[]);

    The initial deformation is given by an Okada fault model with the following parameters. The iterate = adapt option will iterate this initialisation until our adapt() function above returns zero i.e. until the deformations are resolved properly.

        fault (x = 94.57, y = 3.83,
	   depth = 11.4857e3, // fixme: this should be a length, but okada.h does not seem to be dimensionally consistent!!
	   strike = 323, dip = 12, rake = 90,
	   length = 220e3, width = 130e3,
	   U = 18,
	   iterate = adapt);
  }
}

    The 4 other fault segments are triggered at the appropriate times (seconds converted to minutes).

    event fault2 (t = 272./60.) {
  fault (x = 93.90, y = 5.22,
	 depth = 11.4857e3,
	 strike = 348, dip = 12, rake = 90,
	 length = 150e3, width = 130e3,
	 U = 23,
	 iterate = adapt);
}

event fault3 (t = 588./60.)
{
  fault (x = 93.21, y = 7.41,
	 depth = 12.525e3,
	 strike = 338, dip = 12, rake = 90,
	 length = 390e3, width = 120e3,
	 U = 12,
	 iterate = adapt);
}

event fault4 (t = 913./60.)
{
  fault (x = 92.60, y = 9.70,
	 depth = 15.12419e3,
	 strike = 356, dip = 12, rake = 90,
	 length = 150e3, width = 95e3,
	 U = 12,
	 iterate = adapt);
}

event fault5 (t = 1273./60.)
{
  fault (x = 92.87, y = 11.70,
	 depth = 15.12419e3,
	 strike = 10, dip = 12, rake = 90,
	 length = 350e3, width = 95e3,
	 U = 12,
	 iterate = adapt);
}

    Outputs

    At each timestep

    We output simple summary statistics for h and u.x on standard error.

    event logfile (i++) {
  stats s = statsf (h);
  norm n = normf (u.x);
  if (i == 0)
    fprintf (stderr, "t i h.min h.max h.sum u.x.rms u.x.max dt speed tn\n");
  fprintf (stderr, "%g %d %g %g %g %g %g %g %g %ld\n",
	   t, i, s.min, s.max, s.sum, n.rms, n.max, dt, perf.speed, grid->tn);

    We also use a simple implicit scheme to implement quadratic bottom friction i.e. \displaystyle \frac{d\mathbf{u}}{dt} = - C_f|\mathbf{u}|\frac{\mathbf{u}}{h} with C_f=10^{-4}.

      double C_f = 1e-4;
  foreach() {
    double a = h[] < dry ? HUGE : 1. + C_f*dt*norm(u)/h[];
    foreach_dimension()
      u.x[] /= a;

    That is also where we update hmax.

        if (h[] > dry && h[] + zb[] > hmax[])
      hmax[] = h[] + zb[];
  }
}

    Snapshots

    Every 60 minutes, the h, z_b and hmax fields are interpolated bilinearly onto a n x n regular grid and written on standard output.

    event snapshots (t += 60; t <= 600) {

#if !_MPI
  printf ("file: t-%g\n", t);
  output_field ({h, zb, hmax}, stdout, n = 1 << maxlevel, linear = true);
#endif

    We also save a snapshot file we can restart from.

      dump (file = "dump");
}

    After completion of the simulation, doing

    make tsunami/plots

    will generate the inline plots such as this one:

    # We first split the large standard output file into its subfiles (as
# defined by the "file:" keyword, see tsunami.c

! awk '{ if ($1 == "file:") file = $2; else print $0 > file; }' < out

set term pngcairo enhanced size 700,700 font ",8"
set output 'maximum.png'

# this sets the color palette to "jet"
set palette defined ( 0 0 0 0.5647, 0.125 0 0.05882 1, 0.25 0 0.5647 1, \
                          0.375 0.05882 1 0.9333, 0.5 0.5647 1 0.4392, \
                      0.625 1 0.9333 0, 0.75 1 0.4392 0, \
                      0.875 0.9333 0 0, 1 0.498 0 0 )

# here we plot the value of hmax ($5) but only for wet cells 
# (i.e. h = $3 > 1e-3), dry cells take a 'no data' value i.e. 1e1000
# We use only the last output file (i.e. 't-600' minutes = 10 hours)

unset key
set size ratio -1
set xrange [67:105]
set yrange [-10:25]
set pm3d map
set logscale cb
set cbrange [0.1:10]
splot 't-600' u 1:2:($3 > 1e-3 ? $5 : 1e1000)

# we trim the image
! mogrify -trim maximum.png

# we remove the subfiles of tsunami/out
! rm -f t-*
    Maximum wave elevation (metres) reached over 10 hours. (script)

    Maximum wave elevation (metres) reached over 10 hours. (script)

    Movies

    This is done every minute (t++).

    We use the mask option of output_ppm() to mask out the dry topography. Any part of the image for which m[] is negative (i.e. for which etam[] < zb[]) will be masked out.

    event movies (t++) {
  scalar m[], etam[];
  foreach() {
    etam[] = eta[]*(h[] > dry);
    m[] = etam[] - zb[];
  }
  output_ppm (etam, mask = m, min = -2, max = 2, n = 512, linear = true,
	      file = "eta.mp4");

    After completion this will give the following animation

    Animation of the wave elevation. Dark blue is -2 metres and less. Dark red is +2 metres and more.

    We also use the box option to only output a subset of the domain (defined by the lower-left, upper-right coordinates).

      output_ppm (etam, mask = m, min = -2, max = 2, n = 512, linear = true,
	      box = {{91,5},{100,14}}, file = "eta-zoom.mp4");

    Animation of the wave elevation. Dark blue is -2 metres and less. Dark red is +2 metres and more.

    And repeat the operation for the level of refinement…

      scalar l = etam;
  foreach()
    l[] = level;
  output_ppm (l, min = MINLEVEL, max = maxlevel, n = 512, file = "level.mp4");

    Animation of the level of refinement. Dark blue is 5 and dark red is 10.

    …and for the process id for parallel runs.

    #if _OPENMP || _MPI
  foreach()
    etam[] = tid();
  double tmax = npe() - 1;
  output_ppm (etam, max = tmax, n = 512, file = "pid.mp4");
#endif // _OPENMP || _MPI
}

    Animation of the OpenMP process id.

    Tide gauges

    We define a list of file names, locations and descriptions and use the output_gauges() function to output timeseries (for each timestep) of \eta for each location.

    Gauge gauges[] = {
  // file   lon      lat         description
  {"coco", 96.88,  -12.13, "Cocos Islands, Australia"},
  {"colo", 79.83,    6.93, "Colombo, Sri Lanka"},
  {"hani", 73.18,    6.77, "Hanimaadhoo, Maldives"},
  {"male", 73.52,    4.18, "Male, Maldives"},
  {"gana", 73.17,   -0.68, "Gan, Maldives"},
  {"dieg", 72.38,    -7.3, "Diego Garcia, UK"},
  {"rodr", 63.42,  -19.67, "Rodriguez I., Mauritius"},
  {"loui", 57.5,   -20.15, "Port Louis, Mauritius"},
  {"lare", 55.3,   -20.92, "La Reunion, France"},
  {"hill", 115.73, -31.82, "Hillarys, Australia"},
  {"sala", 54,         17, "Salalah, Oman"},
  {"laru", 55.53,   -4.68, "Pointe La Rue, Seychelles"},
  {"lamu", 40.9,    -2.27, "Lamu, Kenya"},
  {"zanz", 39.18,   -6.15, "Zanzibar, Tanzania"},
  {"chen", 80.3,     13.1, "Chennai, India"},
  {"para", 86.7,    20.26, "Paradip, India"},
  {"visa", 83.28,   17.68, "Visakhapatnam, India"},
  {"koch", 76.26,    9.96, "Kochi, India"},
  {"morm", 73.8,    15.42, "Mormugao, India"},
  {"okha", 69.08,   22.47, "Okha, India"},
  {"tuti", 78.15,     8.8, "Tuticorin, India"},
  {"taru", 99.65,   6.702, "Tarutao, Thailand"},
  {"tapa", 98.425,  7.765, "Tapaonoi, Thailand"},
  {NULL}
};

event gauges1 (i++) output_gauges (gauges, {eta});

    As before gnuplot processes these files to produce this image:

    reset
set term svg enhanced size 625,800 font ",10"
set multiplot layout 5,1 scale 1,1.1
set xrange [3:8]
set key bottom right
set title 'Hanimaadhoo, Maldives'
plot 'hani' u ($1/60.):2 w l t 'modelled', \
     '../hanires.txt' u 1:($2/100.) w lp t 'observed'
unset key
set title 'Male, Maldives'
plot 'male' u ($1/60.):2 w l t 'modelled', \
     '../maleres.txt' u 1:($2/100.) w lp t 'observed'
set title 'Gan, Maldives'
plot 'gana' u ($1/60.):2 w l t 'modelled', \
     '../ganares.txt' u 1:($2/100.) w lp t 'observed'
set title 'Diego Garcia'
plot 'dieg' u ($1/60.):2 w l t 'modelled', \
     '../diegres.txt' u 1:($2/100.) w lp t 'observed'
set title 'Columbo, Sri Lanka'
set xrange [2.5:8]
plot 'colo' u ($1/60.):2 w l t 'modelled', \
     '../colores.txt' u 1:($2/100.) w lp t 'observed'
unset multiplot
    Comparison between observed and simulated timeseries (hours) of wave elevations (metres) for a selection of tide gauges. (script)

    Comparison between observed and simulated timeseries (hours) of wave elevations (metres) for a selection of tide gauges. (script)

    Google Earth KML file

    We also generate images and a Keyhole Markup Language file which can be imported into Google Earth to superpose the evolving wave height field on top of Google Earth data.

    event kml (t += 15)
{
  static FILE * fp = fopen ("eta.kml", "w");
  if (t == 0)
    fprintf (fp,
	     "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n"
	     "<kml xmlns=\"http://www.opengis.net/kml/2.2\">\n"
	     "  <Folder>\n");
  fprintf (fp,
	   "    <GroundOverlay>\n"
	   "      <TimeSpan>\n"
	   "        <begin>2004-12-26T%02d:%02d:00</begin>\n"
	   "        <end>2004-12-26T%02d:%02d:00</end>\n"
	   "      </TimeSpan>\n"
	   "      <Icon>\n"
	   "	    <href>eta-%g.png</href>\n"
	   "      </Icon>\n"
	   "      <LatLonBox>\n"
	   "	    <north>35</north>\n"
	   "	    <south>-19</south>\n"
	   "	    <east>121</east>\n"
	   "	    <west>67</west>\n"
	   "      </LatLonBox>\n"
	   "    </GroundOverlay>\n",
	   8 + ((int)t)/60, ((int)t)%60,
	   8 + ((int)t + 15)/60, ((int)t + 15)%60,
	   t);
  fflush (fp);
  scalar m[], etam[];
  foreach() {
    etam[] = eta[]*(h[] > dry);
    m[] = etam[] - zb[];
  }
  char name[80];
  sprintf (name, "eta-%g.png", t);
  output_ppm (etam, file = name, mask = m,
	      min = -2, max = 2, n = 1 << maxlevel, linear = true);
  if (t == 600)
    fprintf (fp, "</Folder></kml>\n");
}

    Adaptivity

    And finally we apply our adapt() function at every timestep.

    event do_adapt (i++) adapt();