Environmental Engineering Reference
In-Depth Information
Fig. 5.2 Top-level structure for multi-model operational forecast system. Multiple wind forecasts
are used to drive sets of hydrodynamic models using different wave model parameters and different
hydrodynamic parameters. Highlighted section is expanded in Fig. 5.3
full Monte-Carlo methods (random selection of the parameters and conditions over
a statistically valid set at each model level) would produce an impractically-large set
of simulations. Selection of high and low conditions/parameters requires thorough
understanding of uncertainty contributions, which can be readily handled by hindcast
analyses during development of an operational system(Sect. 5.8 ). Asystemwith three
condition sets at each level (i.e. Figs. 5.2 and 5.3 ) would require 27 hydrodynamic
model runs and 243 oil spill model runs, which could be accomplished with modest
investment in a set of standard workstation computers. A relatively small number of
perturbations in each step of the modelling provides a wide range of model results
for statistical processing.
Implementing an operational system with this large number of simulations might
appear to be computationally impractical. However, the continuing advance of low-
cost multiprocessor workstations and GPU computing changes the question from
“can we do this?” to “given our budget, how big can N x be for each component?”
Compared to hydrodynamic models, oil spill models run very quickly and take rela-
tively little memory, so running a large number of such models is eminently practical.
Search WWH ::




Custom Search