Tools for applying adaptive mesh refinement and automated tuning to wave model simulations
Richard M. Gorman, Henrique Alves
Ntl Institute of Water & Atmospheric Research, NZ,
EMC
30 Nov, 2:30 pm, in 2155
Abstract:
This presentation will cover two areas of research undertaken during a
sabbatical visit to NCEP. The first involves development of an
adaptive-grid version of WAVEWATCH III. We will also discuss
development and application of an optimisation suite (“Cyclops”) based
on the Cylc workflow engine, and its application to improve the skill
of NWPs.
A fully adaptive quadtree grid has been added to the WAVEWATCH III
model package, to dynamically vary spatial resolution to match the
requirements of the evolving solution. This enables it to give the
finest resolution where wave energy density has the strongest variation
(e.g. in the vicinity of an intense, moving storm system), while only
using lower resolution elsewhere. This has the potential to
significantly improve the trade-off between model accuracy and
simulation time.
The optimisation suite Cyclops, based on the Cylc workflow engine
(http://cylc.github.io/cylc/), implements a wide selection of
optimisation algorithms to calibrate any modelling system that has
itself been implemented as a (separate) Cylc model suite. Cyclops is
designed so that the separate optimisation and model suites only need
to be set up to exchange parameter values and computed error metrics as
simple file I/O. This makes it relatively straightforward to apply the
optimisation suite to calibrate a modelling system that has already
been implemented in an existing Cylc suite.
Results from applications of Cyclops to calibrate wind-wave hindcasts
using the WAVEWATCH III model are presented for simulations made using
ECMWF and NCEP atmospheric forcing. For the latter case, significant
wave model skill-score improvements are obtained when using the spatial
average of the root-mean-square error of significant wave height,
relative to collocated altimeter records, as the cost function guiding
a nonlinear optimization algorithm within Cyclops.