Computational Penalties of Component Based Models: An Urban Stormwater Component-Based Modeling Application Using Open MI
Location
Eccles Conference Center Auditorium
Event Website
http://water.usu.edu
Start Date
3-31-2015 8:10 AM
End Date
3-31-2014 12:00 AM
Description
Component-based environmental modeling, or loose model coupling, has been proposed as an alternative to traditional tight model coupling, which is characterized by inflexible and often large model codes with highly interdependent functions compiled into a single execution unit. Loosely coupling model components developed by decomposing complex systems into smaller or less complex independent units promises earth systems modelers: (1) an approach to better explore feedbacks between domains of different disciplines that are typically modeled independently, and (2) a way to experiment with different process formulations to select those that are most appropriate for a particular application. The additional function calls, data transformations, and discontinuity at the connection points between model components resulting from the use of component-based modeling may, however, give rise to computational penalties, including increased simulation time and mass balance error. In the study presented here, we sought to investigate these computational penalties as the number of coupled model components increases using the Open Modeling Interface (OpenMI), which is component-based modeling interface specification. A Stormwater Management Model (SWMM) application developed for the City of Logan, Utah, USA and run in its standard, tightly coupled configuration served as a reference against which several configurations of coupled OpenMI-compliant SWMM model components were compared. The various configurations of coupled OpenMI SWMM model components were derived by decomposing the reference SWMM model first by process (i.e., runoff coupled to routing) and then by space (i.e., catchments or groups of catchments coupled together). Results showed that simulation times increased linearly as the number of connections between model components increased. The results also showed that changes in total mass balance error introduced through coupling were dependent on how well each model component was able to resolve the time series data it received. This study also demonstrates and proposes some strategies to address these computational penalties in component-based modeling frameworks.
Computational Penalties of Component Based Models: An Urban Stormwater Component-Based Modeling Application Using Open MI
Eccles Conference Center Auditorium
Component-based environmental modeling, or loose model coupling, has been proposed as an alternative to traditional tight model coupling, which is characterized by inflexible and often large model codes with highly interdependent functions compiled into a single execution unit. Loosely coupling model components developed by decomposing complex systems into smaller or less complex independent units promises earth systems modelers: (1) an approach to better explore feedbacks between domains of different disciplines that are typically modeled independently, and (2) a way to experiment with different process formulations to select those that are most appropriate for a particular application. The additional function calls, data transformations, and discontinuity at the connection points between model components resulting from the use of component-based modeling may, however, give rise to computational penalties, including increased simulation time and mass balance error. In the study presented here, we sought to investigate these computational penalties as the number of coupled model components increases using the Open Modeling Interface (OpenMI), which is component-based modeling interface specification. A Stormwater Management Model (SWMM) application developed for the City of Logan, Utah, USA and run in its standard, tightly coupled configuration served as a reference against which several configurations of coupled OpenMI-compliant SWMM model components were compared. The various configurations of coupled OpenMI SWMM model components were derived by decomposing the reference SWMM model first by process (i.e., runoff coupled to routing) and then by space (i.e., catchments or groups of catchments coupled together). Results showed that simulation times increased linearly as the number of connections between model components increased. The results also showed that changes in total mass balance error introduced through coupling were dependent on how well each model component was able to resolve the time series data it received. This study also demonstrates and proposes some strategies to address these computational penalties in component-based modeling frameworks.
https://digitalcommons.usu.edu/runoff/2015/2015Posters/38