Operational Oil Spill Modelling: From Science to Engineering Applications in the Presence of Uncertainty - Mathematical Modelling and Numerical Simulation of Oil Pollution Problems - page 109

on a regular basis, which requires a dynamic modelling system; i.e. the operational
system should automatically obtain newdata/forecasts, integrate the newdata into the
models, and re-run all the models to produce a new set of forecasts and visualizations.
Techniques for handling these issues are discussed in Sect. 5.7 , below. Tomake such a
system practical, computational power needs to be made available such that an entire
multi-model forecast sequence, from wind to hydrodynamics to oil spill model, can
be completed before the next COS data and wind forecasts become available.
A further difficulty in providing operational forecasts is the fact that any Lagrang-
ian particle simulation has a limited time-horizon for reliability. Because the
Lagrangian particles are inherently integrative of error, their divergence from the
real-world will increase with time. The most effective operational system will inte-
grate data sources for estimating the real-world position of the oils spill (e.g. through
satellite tracking, [ 54 ]) that can be used to periodically reset the oil spill particles to
a new “known” position.
Clearly, creating and automating a multi-model operational system with forecast
uncertainty presents a number of challenges, including (i) generation of perturbed
wind forecasts, (ii) selecting parameter sets for wave, hydrodynamics, and oil spill
models, (iii) selecting sets of reasonable initial conditions for the oil spill, (iv) ana-
lyzing and visualizing the combined forecast data, (v) automated updating of models
as new data and forecasts are received, and (vi) creating a system that integrates
models and data so that the user provides the location, estimated size, and oil type
that is spilled and obtains an animation of the time-evolution of a probability map
for the oil spill. To address some of these challenges, authors Sayol and Orfila have
developed new techniques for probability simulations within oil spill models and
probability mapping visualization (Sect. 5.9 ), while simultaneously authors Hou and
Hodges have developed the HyosPy system of model integration (Sect. 5.7 ).
5.7 Multi-Model Integration and Updating Predictions
The Hydrodynamic and oil spill Python (HyosPy) code has been developed as a
testbed for integrating hydrodynamics and oil spill models in a flexible manner
[ 18 , 19 ]. Presently, HyosPy is designed to integrate COS data, wind forecasts, and
multiple hydrodynamic models linked to independent oil spill models, as illustrated
in Fig. 5.4 . The results are visualized in Google Earth/Maps applications. HyosPy
uses the Python scripting language, which provides a flexible “wrapper” to integrate
existing models, servers, connections to online data services, and visualization tools.
The present version of HyosPy is being tested for coastal embayments of Texas (USA)
with automatic tidal data downloads from the Texas Coastal Ocean Observation
Network (TCOON) [ 49 ] and wind forecast data from a Texas A&M University
server. The hydrodynamic model used is SELFE [ 42 ], which has been under review
by the Texas Water Development Board (TWDB) and General Land Office (TGLO)
as a replacement for the TxBlend model, which is presently the operational oil spill
model used inside barrier islands along the Texas coastline [ 52 ]. The oil spill model