 |
| Titel |
Development of an integrated modelling framework: comparing client-server and demand-driven control flow for model execution |
| VerfasserIn |
Oliver Schmitz, Derek Karssenberg, Kor de Jong, Jean-Luc de Kok, Steven M. de Jong |
| Konferenz |
EGU General Assembly 2014
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 16 (2014) |
| Datensatznummer |
250097299
|
| Publikation (Nr.) |
 EGU/EGU2014-12863.pdf |
|
|
|
|
|
| Zusammenfassung |
| The construction of hydrological models at the catchment or global scale depends on the
integration of component models representing various environmental processes, often
operating at different spatial and temporal discretisations. A flexible construction of
spatio-temporal model components, a means to specify aggregation or disaggregation to
bridge discretisation discrepancies, ease of coupling these into complex integrated models,
and support for stochastic modelling and the assessment of model outputs are the desired
functionalities for the development of integrated models. These functionalities are
preferably combined into one modelling framework such that domain specialists can
perform exploratory model development without the need to change their working
environment.
We implemented an integrated modelling framework in the Python programming
language, providing support for 1) model construction and 2) model execution. The
framework enables modellers to represent spatio-temporal processes or to specify
spatio-temporal (dis)aggregation with map algebra operations provided by the PCRaster
library. Model algebra operations can be used by the modeller to specify the exchange of data
and therefore the coupling of components. The framework determines the control flow for the
ordered execution based on the time steps and couplings of the model components given by
the modeller. We implemented two different control flow mechanisms. First, a client-server
approach is used with a central entity controlling the execution of the component models and
steering the data exchange. Second, a demand-driven approach is used that triggers the
execution of a component model when data is requested by a coupled component
model.
We show that both control flow mechanisms allow for the execution of stochastic,
multi-scale integrated models. We examine the implications of each control flow
mechanism on the terminology used by the modeller to specify integrated models, and
illustrate the applicability of both approaches by constructing integrated models
from components simulating hydrological processes, land use change or biomass
growth.
The PCRaster software runs on Microsoft Windows, Linux and OSX. More information
about the integrated modelling prototypes and download of the PCRaster software collection
is available at http://www.pcraster.eu/. |
|
|
|
|
|
|