Professional Documents
Culture Documents
The Earth by TELEMAC: Sébastien E. Bourban, Michael S. Turnbull and Alan J. Cooper
The Earth by TELEMAC: Sébastien E. Bourban, Michael S. Turnbull and Alan J. Cooper
1
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
Subsequent development phases include the application A. A scalable unstructured mesh of the Earth
of the Earth by TELEMAC to predicting internal tides A data source often used to define the shoreline of coastal
(resulting from the combination of stratified waters and waters is the Global Self-consistent, Hierarchical, High-
tides), computing mean dynamic topography (caused by resolution Shoreline (GSHHS) database. GSHHS is a high-
temperature variations) and to forecasting ocean currents. resolution shoreline dataset that has undergone extensive
The second objective of the work presented here is to processing and is said to be free of internal inconsistencies
open up research and collaboration with national and (see [1]). It can be accessed from www.ngdc.noaa.gov or
international organisations, from around the world, to from www.soest.hawaii.edu/pwessel/gshhs.
enhance the Earth by TELEMAC model with further Based on these shorelines, mesh generators can generate
knowledge, expertise and data. Already, separate research the unstructured mesh given a constant or variable resolution
and development activities carried out by HR Wallingford in (triangle edge-length). At this stage, a constant mesh
collaboration with renowned universities will be combined resolution is assumed, the value of which will depend on the
with this project to extend its capabilities. For instance, work computing resources available.
done on Algorithmic Differentiation will be combined to
extend the Earth by TELEMAC to possible data assimilation. Unfortunately, currently available mesh generators are
Similarly, work done on Ice Modelling will help with the based on iterative methods that are too slow for extremely
characterisation of the influence of the cold regions. large number of vertices. Besides, most generator cannot deal
with a round Earth, i.e. the fact that 0 deg = 360 deg.
The solution was, therefore, not to generate the mesh but
II. SETUP OF THE EARTH BY TELEMAC to directly compute it, ignoring for a while the complication
The purpose of the model is to represent the entire Earth, due to the presence of landmasses. Working on a full sphere,
more specifically its water bodies, at a higher resolution than the main idea was to start from a known triangularly facetted
the current state of the art, below the kilometre if possible. 3D shape and to recursively refine it using the existing mesh
The first step in setting up a model with TELEMAC is to refinement tools available within the python scripts of the
setup an unstructured mesh of the computation domain as TELEMAC system: the icosahedron, a 3D shape of 20
defined by its extent and its resolution. The second step in triangles (elements), was chosen.
setting up a model is to map the bathymetry. The third step is Fig. 1 below shows the initial icosahedron rounding up
to define its temporally and spatially varying open boundary the contours of the Earth, the first refinement (splitting each
conditions. Luckily, the advantage of modelling the whole triangle into 4 smaller triangles) together with the re-
Earth is that there are no open boundaries! projection of the new vertices on the sphere, and the
subsequent refinement from iteration 2 through iteration 8.
… below
1 km resolution
2
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
If we start with the 360 degrees of the circumference of the of global tidal harmonic dataset TPXO (see [20]), or
sphere based on a circle going through the poles, the the global 3D oceanic model HyCOM (see [21]).
icosahedron splits the sphere in 6 divisions, i.e. of 60 degrees
for each foot of the triangle. The edge-length of a triangle is, The highest resolution used so far to a compute an
therefore, slightly larger. If we now look at the circumference unstructured mesh for TELEMAC has been for
of the sphere based on a circle going along the equator after the Iteration 12. Higher iterations would require slight
first refinement, icosahedron splits the sphere in 10 divisions, modifications to the existing python scripts to include
i.e. 36 degrees for the first refinement, or 72 degrees for the long integers, as well as heavy modifications to the
icosahedron (5 divisions). In fact, depending on which pair of TELEMAC source code to allow long integer in
nodes, the edge length of the first refinement is either 31.72 or appropriate places.
36 degrees. Computation of the mesh for Iteration 11 takes about
Let us chosen the longest of the edge lengths (i.e. 10 minutes (on a standard computer) and 45 minutes
72 degrees), approximating the radius of the Earth, R, by for Iteration 12. This remains acceptable and far more
6,371,000 m, the circumference is 2πR = 40,030,173 m. With efficient than generating the mesh. Iteration 13 (below
that said, the “edge length” (distance when projected on the the kilometre resolution) should not take more than
surface of the sphere) of the icosahedron is 4 hours, although this was not tested.
(2πR/5) = 8,006,035 km. The chosen methodology (based on subsequent
Subsequent refinements of the icosahedron multiply the refinements) to compute the unstructured mesh of the
number of elements by four while halving the edge length Earth could be an advantage in the future to support
(following the curvature the sphere). TABLE 1 below shows the multi-grid methods in order to accelerate further
resulting metrics for subsequent mesh refinements. solutions for the TELEMAC system (although not part
of the current research project). In any case, the
TABLE 1 REFINEMENTS OF THE ISOCAHEDRON approach taken has been beneficial to the project as
testing of functionalities can be done on coarser
Iterations Unstructured mesh characteristics meshes, while developing final results are based on the
(sphere) 360 (deg) 40,030,173 (m) (elements) finer mesh.
isocahedron 72 8,006,034.7 20
1 36 4,003,017.4 80
B. Mapping Bathymetry and Topography
2 18 2,001,508.7 320 Having created an unstructured mesh of the Earth, the
3 9 1,000,754.3 1,280 bathymetric and topographic datasets are then mapped using a
trivial 3-point weighted interpolation method.
4 4.5 500,377.2 5,120
Using the software Blue Kenue (see [4]) Fig. 2 below
5 2.25 250,188.6 20,480 shows a coloured rendering of the resulting mesh, with mapped
6 1.125 125,094.3 81,920 values of bathymetry (in shades of blues) and topography (in
shades of green).
7 ~1/2 62,547.1 327,680
10 ~ 1 / 14 7,818.4 20,971,520
11 ~ 1 / 28 3,909.2 83,886,080 a
12 ~ 1 / 57 1,954.6 335,544,320
3
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
The GEBCO dataset was selected at this stage (see [2]). It been removed (the Pacific Ocean is visible through the Asian
has a resolution of about 1/16th of a degree. As the Earth by continent, South America through the African continent, …).
TELEMAC is refined further towards the kilometre, other The gridded sphere inside the Earth is only shown for
sources of local bathymetries will have to be assembled for perspective.
accurate results, particularly near the coasts. For instance, the
marine data provider SeaZone (see [3]), can be put to the As previously said, several models of the Earth were
contribution by HR Wallingford. developed in the course of this project, at various resolutions
(mainly between Iteration 9 and 11), in order to effectively
The shades of green shown on Fig. 2 are here, in fact, not develop, test, validate and finally forecast various physical
representing topography but were rather extracted from the aspects of all water bodies.
“Blue Marble” satellite photo of the Earth surface (see [5]).
4
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
width, its dislocation, its strike direction, its dip angle, and its gravitational forces (relative rotations of the Sun, the Moon,
slip (see illustrative Fig. 5 below). the Earth and various corrections in orbit trajectories), a feature
that is already available in the TELEMAC system, using the
keyword:
TIDE GENERATING FORCE
After calibration of the hydrodynamic model based on the
TELEMAC-2D component (to account for 3D phenomenon
such as internal tides), model results of surface levels and
depth average velocities were favourably compared with the
TPXO dataset.
Figure 5. Tsunami characterisation
In order to demonstrate the applicability of TELEMAC-2D
It should be noted that release v7p2 of the TELEMAC to forecast tides over the entire globe, another animation was
system only includes the possibility to define one tsunami created and posted on YouTube (see [22]). For reference, an
source, using the keywords: arbitrary 15-day period starting on January 13, 2017 was
selected. Fig. 6 below shows two insets (two sides of the
OPTION FOR TSUNAMI GENERATION; and Earth), of snapshots of that animation of the Earth by
PHYSICAL CHARACTERISTICS OF THE TSUNAMI TELEMAC. Colours represent the maximum tidal range
It is anticipated, as a result of this internal project, that the computed over a spring tide (with red indicating the highest
keywords above will be extended in release v7p3 to the and blue the lowest variation in tidal range). For comparison
definition of multiple sources, whether combined to define one purposes, Fig. 7 shows the same snapshots but extracted from
event, and / or triggered as separate events at different times. the TPXO dataset.
For reference, the selected events shown in the animation
posted on YouTube were: Solomon Islands 2007 (see [10]),
Tohuku 2011 (see [11]), Sumatra 2004 (see [12]),
Makran/Balochistan 1945 (see [13]), Greece 1956 (see [14]),
Lisbon 1755 (see [15]), Dominican Republic 1946 (see [16]),
Ecuador/Colombia 2016 (see [17]), Valvidia 1960 (see [18])
and Kamchatka 1952 (see [19]).
With a resolution of about 4 km (Iteration 11) and a time
step of 1 minute, the 24 hour simulation was computed in less
than 5 hours on 36 compute cores only. Higher speed up can be
achieved with a higher number of cores, particularly for a Figure 6. Forecasting tides over the Earth by TELEMAC
possible forecasting mode.
B. Forecasting tides
In order to fulfil the first objective of this internal project
(to deliver comprehensive and detailed global modelling
resources to support environmental hydraulics studies
anywhere in the world), the second phase looked at an
important aspect of all coastal models: the forecasting of tidal
information.
Typical of the setup of a coastal (hydrodynamic) model,
Figure 7. Extraction of observed tides from the TPXO dataset
one would extract tidal harmonic information from a global
dataset or from a range of tidal gauge observations and re- While there are differences between the TELEMAC-2D
synthetize a signal on its open boundary for various periods. forecast and the TPXO dataset, these could be explained by the
The TPXO dataset (see [20]) is one such global dataset. It is absence 3D processes. It is expected that TELEMAC-3D, with
widely used in earth sciences and was first integrated within an appropriate representation of ocean currents and internal
the TELEMAC system in 2011, by the authors of this article. It waves / tides as well as a self-attracting force would provide a
provides several harmonic constituents computed from a best- fairer comparison (see Chapter V).
fit of tidal levels measured along remote sensing tracks from
the TOPEX/POSEIDON and its subsequent JASON satellite C. Forecasting storm surges
programs. In the continuation of Phases 1 and 2, the third phase of the
project looked at atmospheric drivers, such as winds and
As previously noted, the Earth by TELEMAC has no open atmospheric pressure. Atmospheric temperature and other
boundaries. Therefore, the forecasting of tides could not have aspects of the thermal balance would be tested later, before
been imposed but rather directly computed from known going 3D. Atmospheric drivers (whether drivers of the
5
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
hydrodynamic or the heat budget) are existing features of the Again, the Earth by TELEMAC has no open boundaries.
TELEMAC system and are often used on a wide range of Therefore, the forecasting of waves has to be the sole results of
consultancy studies, particularly in metocean studies. winds. A simulation of the TOMAWAC component was,
therefore, setup for the Earth by TELEMAC, with waves
Therefore, similarly to the application of the Earth by driven only by temporally and spatially varying winds. Winds
TELEMAC to the propagation of tsunami waves, the were extracted from the ERA-5 dataset as well as waves for
applicability of TELEMAC-2D to predicting storm surges over comparison purposes.
the entire globe was simply tested and demonstrated through
another animation posted on YouTube (see [23]). In order to demonstrate the applicability of TOMAWAC to
forecast waves over the entire globe, another animation was
The animation shows water surges (ranging from red: created and should be posted on YouTube in early November
highs, down to purple: lows) driven by storms (shown by wind 2017. For reference, an arbitrary 60-day period starting on
vectors) running across the globe for 11 known storms within November 1, 2016 was selected. Fig. 9 below shows two insets
an 85-day sequence. Fig. 8 below shows two insets (two sides (two sides of the Earth), of snapshots of that animation of the
of the Earth), of snapshots of that animation. Note that the Earth by TELEMAC. Colours represent the wave heights
winds are shown several hundred kilometres above the surface computed (with red indicating the highest waves blue the
so that the surge and wind field can be seen more readily. lowest waves). For comparison purposes, Fig. 10 shows the
same snapshots but extracted from the ERA-5 dataset.
6
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017
7
24th Telemac-Mascaret User Conference Graz, Austria, 17-20 October, 2017