EH Workbench Motivation
--Michael Tobis, 1999
The central problem in the physics of the environment is the coupling of systems at multiple scales. Typically, the various scales and phenomena have been studied by different disciplines and subdisciplines. As understanding progresses, interest develops in phenomena that couple principles discovered in separate disciplines. For example, the El Nino phenomenon can only be understood by combining ideas from meteorology and oceanography, and can only be quantitatively addressed by models including quantitative representations of atmosphere and ocean.
Similarly, in addressing the impacts of climate and climate change on regional systems, ways have to be found to include quantitative models of both the climate system and the impacted environmental phenomena. The particular example addressed by this project is the effect of climate change on severe flooding events.
While it is possible to build a coupled model from a blank slate, often this is unrealistically expensive. Therefore, the quantitative models are frequently accomplished by coupling pre-existing models of the relevant subsystems.
A problem often mentioned by professionals using large models is the difficulties inherent in getting them to work properly. These problems are more than doubled when two or more pre-existing models need to be run to investigate a particular scientific issue.
Commercially important fields which involve high performance computing have made some progress toward insulating the domain professionals from the details of setting up the computation. Tools to accomplish such tasks, which in principle are highly automatable, are called "problem solving environments". Our goal is to bring such an environment to quantitative studies in the environmental sciences.
In addition to the more or less seamless integration of component models into a single functional environment, the problem solving environment must provide feedback to the user, usually through visualization tools. We have built our visualization tools around the VisAD library. VisAD is a Java-based API that itself makes extensive use of the standard java extension Swing API for GUI construction. Therefore Java and Swing provided a natural basis for the user interface.
Finally, using Globus technology, we implemented methods for allocating and instantiating computations remotely, moving toward seamless allocation. There is also a less capable version of the workbench which uses secure shell ("ssh") invocation methods, with its smaller capability being traded of for a rather simpler setup.
Our group also produced a virtual reality visualization of a coupled model run, using the modified Cave5D software. The Java based VisAD workbench, the coupled models, and the virtual reality visualization were all demonstrated at SC98 in Orlando in November of 1998.