Environmental Engineering Reference
In-Depth Information
any operation involving simulations, such as simulation
per se
or finalizing parameter
calibration. This is not because of the component based structure of APES, but
rather because complex systems are being simulated with process-based models.
Automated optimization procedures in model chains may produce results which are
meaningless in biophysical terms. Such simulation anomalies due to either
inappropriate input data or even the misuse of the simulation model might be
evident working with the biophysical system simulation alone, but a misuse of
APES (or of any other process based system) in model chains may be very difficult
to spot and could have an unpredictable impact on final results of the analysis.
When included in a model chain, it is advisable to link APES simulations to
other models asynchronously, allowing for simulation results to be evaluated by
an analyst prior to using APES outputs as inputs for further processing.
The paragraph above is not meant to suggest that a complex simulation system
should not be integrated in model chains such as the SEAMLESS integrated frame-
work. Instead, it is meant to stress the importance of implementing procedures to
facilitate the evaluation of intermediate results, both by domain experts and via
specific utilities.
Concluding Remarks
APES development represents a paradigm shift for two reasons. First, APES is not
proposed as “the” model. Instead it stresses the need for broadening modelling
approaches and for comparing them at a finer granularity than for whole simulation
systems. Secondly, compared to the first modelling frameworks for overcoming the
problems of monolithic models, APES moves the focus onto components and their
re-usability outside APES itself, even as stand-alone entities. A somewhat surprising
result emerged during the initial development of APES. Contrary to past experience
when implementation of complex systems has often been the most challenging
task, the major difficulty has turned out to be thinking in “modular”, “multiple
choices”, “transparent” modelling terms. The goal of making models available as
discrete, re-usable units aimed at including ideally one process in each basic model
unit has forced us to thoroughly analyze assumptions and the independence of each
model from others. In fact, developing model components, even with the requirements
listed in the previous paragraphs, is a modest challenge in terms of implementation,
but it forces us to formalize modelling knowledge and the problem of model linkage
and re-use. Technology is expected to move more and more towards declarative
modelling in an operational way. The work carried out in creating fine-granularity,
discrete model units, encapsulating a semantically-rich description of interfaces, has
involved discussing and advancing understanding of many aspects of model
assumptions and structures, and will be of great help in that direction.
APES development during the SEAMLESS project has led to an increasing
opportunity to concentrate on modelling options by re-using expertise in different
domains. APES is offered as a complete simulation tool, but also, and of no less
