Environmental Engineering Reference
In-Depth Information
retrieves the results, calculates the objective function to be minimized and deter-
mines the parameter values for the next iteration.
The current version of the PE has five different algorithms for parameter estimation.
The different parameter estimation methods correspond to different hypotheses about
model error.
The Sensitivity Analysis Component
Sensitivity analysis (SA) is a fundamental tool in the building, use and understanding
of all types of mathematical model. SA provides information regarding the
behaviour of the underlying simulated system. This information ranges from the
identification of relevant model factors (parameters or inputs) to model reduction or
simplification, better understanding of the model structure for given components of
a system, model quality assurance, and model building in general. Among the most
commonly used methods, it is possible to identify three classes: screening methods,
regression-based methods, variance-based methods. The most used screening
method is the one proposed by Morris, which is particularly effective in identifying
the few important factors in models with many factors or with high computational
requirements. The second class includes the regression methods, which are based on
the computation of standard or partial regression coefficients quantifying the effects
due to a change in a factor value while the others are kept constant. Within this class,
different methods can be used to generate the sample of factors combinations neces-
sary to obtain the model evaluations and therefore to calculate the regression coef-
ficients; here, Latin Hypercube Sampling (LHS), Random, and Quasi-Random
LpTau will be used. The last class, variance-based methods, includes the Fourier
Amplitude Sensitivity Test (FAST), its evolution Extended FAST (E-FAST), and the
method of Sobol'. All the methods belonging to this class compute total sensitivity
indices for first and higher order effects and are demanding in terms of computa-
tional time because of the high number of model evaluations needed for each model
factor. SensitivityAnalysis is a component developed by JRC and CRA (Donatelli
et al., 2009c) with the goal of making available the sensitivity analysis models
implemented in the SimLab library (Saltelli et al. 2004) via a user friendly applica-
tion programming interface, in the memory managed environment of the Microsoft.
NET platform (the Simlab library is available for C, C++, Matlab and Fortran). The
component allows sensitivity analysis to be run on a model of choice using the
methods mentioned above. It is implemented using C# under the Microsoft .NET v
3.0 platform. Sample applications inclusive of source code are provided to allow an
easy start to SensitivityAnalysis use via different software clients.
Remarks on APES Integration in Larger Systems
APES integration, although technically possible at even closer levels than the ones
used in the integration into the SEAMLESS integrated framework should, however,
