This directory contains files for the prototype study for stochasticity conducted by Suresh Kumar of ACFR and Ole Nielsen of GA.
Project outline:
The case study is based on a validation conducted against a wavetank experiment emulating the 1993 Okushiri Island tsunami.
1)Investigate independent, identically distributed errors in bathymetry. Decide a standard deviation of the bathymetry errors.
The important files are:
project.py: Controls number of samples etc
create_realisatons.py: Makes a set of perturbed bathymetries
run_model.py: Run the simulation for each realisation and output three timeseries for each
(2)Establish that the Monte-Carlo estimates have converged
Brute force validation where we run say 1000 simulations and then
2000, and evaluate whether the mean solution at the 3 monitoring
stations has significantly changed.
(3)Measures of performance at the three channels where measurements
are available
(a) 1 standard deviation from the mean (time averaged)
(b) Windowed RMS error measure
(c) Raw RMS error
(d)Visual inspection
Milestone: Evaluate the effects of the stochasticity of bathymetry on
the computed solution using the measures of performance
listed here to decide on continuance of this line of work.
Second mini-project (Pending success of earlier project):
Introduce spatially correlated errors in bathymetry and perform
Monte-Carlo simulations with sparse, locally correlated deviates.
Research Questions:
* Which resolution is necessary?
* What is the effect of noise in the friction terms?
* Effect of Boundary conditions, source terms and initial condition.
* Can the whole process be done stochastically -
perhaps a hybrid where CFD kick starts a stochastic
predictive process (like ITO)