 |
| Titel |
Constructing integrated models: a scheduler to execute coupled components |
| VerfasserIn |
Oliver Schmitz, Derek Karssenberg, Kor de Jong, Jean-Luc de Kok |
| Konferenz |
EGU General Assembly 2011
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 13 (2011) |
| Datensatznummer |
250054215
|
|
|
|
|
|
| Zusammenfassung |
| In integrated modelling, multiple interacting components of various domains from
environmental, social and economic systems are assembled to model complex system
behaviour. The integrated model thereby serves as a means to increase the scientific
understanding as well as a base for policy relevant research. Tools for integrated modelling
need to support construction and coupling of model components, and preferably assistance to
assess models. However, present software frameworks are either tailored to component
construction or to component coupling, whereas a consolidated environment is desired.
Existing frameworks also require profound knowledge in system programming
languages, and offer limited generic support for model assessment. As a result, emphasis
in the model development is currently on construction rather than on testing and
assessment.
To resolve the addressed technical hassle in the model construction process, we propose a
layered software architecture for straightforward construction, coupling, execution and
analysis of model components. Here, we primarily focus on the treatment of temporal
interactions between components. We strive to support different modelling scenarios such as
a combination of components holding continuous discrete time steps, for example
in field-based modelling, and event-based components used in individual based
modelling.
The layered architecture of the software framework is designed as follows: in the first
layer, the model description layer, a model developer constructs components with the help of
spatial and temporal building blocks, and describes the component relationships. By
following the framework guidelines, components can consist of fixed or variable
time step lengths, or confined lifetime within the model run. Also, adapter that
aggregate over time can be specified. These adapter can be used for example to average
output of a component with a daily time step in order to feed a component with a
monthly time step. In the second, the schedule generation layer, the information of
components and their relationships are processed to determine the order of component
execution. The third, the execution layer, organises the actuating of components
and adapter as well as the accounting of component results for post-run analysis
purposes.
We provide the functionality of the framework as modules for the high-level scripting
language Python. Therefore, the model builder is able to use a platform-independent
development environment with a strong support for scientific computing. Moreover,
non-software developers are able to conduct exploratory model construction and analysis
after a short familiarisation phase.
We present a prototype version of the software framework in connection with
spatio-temporal operations from the PCRaster library. We demonstrate how a model
developer can apply the framework to construct complex models with diverse spatial and
temporal discretisations. A case study is used to illustrate the interaction of components with
different time step lengths and process representations. |
|
|
|
|
|
|