OR/14/022 Comparison of approaches

Barkwith A K A P, Pachocka M, Watson C, Hughes A G. 2014. Couplers for linking environmental models: Scoping study and potential next steps. British Geological Survey Internal Report, OR/14/022.

Table 1 is a first attempt at a high level summary of selected coupling technologies which serves as a means to quickly compare some of the key features associated with couplers. Dunlap et al., (2013) describe an approach to assessing coupler features through feature analysis and the creation of feature diagrams, this approach may be considered as a subsequent, more detailed, analysis was required. Carrying out a feature analysis based approach would be much easier to achieve after the couplers described in the following matrix have been whittled down to a short list of candidates.

The technologies compared in the matrix were specifically identified as being relevant, or potentially relevant, to current BGS activities.

Table 1 Comparison of coupling approaches

		CSDMS 1.0	CSDMS 2.0	OpenMI 1.4	OpenMI 2.0	Trident2	CESM-CPL 7	OASIS3-MCT_2.0	FLUME
Background	Open Source	Yes	Yes	Yes	Yes	Yes Community led since 2013, previously an MS initiative	Yes Subject to the IPR rules of embedded software	Yes (LGPL)	No
	Primary research community	Surface dynamics	Surface dynamics + ?	Water	Water + extra env. disciplines	Oceanography	Climate	Climate	Climate
	Central model repository	Yes CSDMS portal	Yes CSDMS portal	Yes but optional FluidEarth	? Could use FluidEarth but no 2.0 models there yet	Some models held on a site called my Experiment and CSIRO have their own repository	No	No	?
Functionality and implementation details	Visual workflow configuration interface	Yes (CMT)	Yes (CMT)	Yes (FluidEarth)	Yes (FluidEarth)	Yes	Yes (GUI)	Yes (GUI)	Yes (GUI)
	Visual ‘programming’ interface Tools for creating model components that require relatively low level of programming experience, recommended by Gou D et al.,, 2012	? None identified	? None identified	Partially 3rd party tools e.g. Visual Studio	Partially 3rd party tools e.g. Visual Studio	Yes	? None identified	? None identified	? None identified
	Programming language for the framework SDK/wrapper code	BMI functions can be written in C, C++, Fortran (all years), Java and Python. CMT conversion done by CSDMS staff	Same as 1.0 except for the documentation no longer mentions Fortran and Java wrapping tool not available yet	C# Java? And supported by XML	C# Java? And supported by XML	.NET (C# & VB.NET)	Fortran	Fortran 77, Fortran 90 and C	?
	Model Languages supported This is a list of languages that pre-compiled models/components can be written in.	C, Fortran (77, 95, 2003), C++, Java, Python	C, Fortran (77, 95, 2003), C++, Java, Python	C# Java, C, C++ Fortran, Pascal (And via 3rd party SSW: MATLAB. Scilab Python)	C# Java, C, C++ Fortran, Pascal (SSW for 2.0 planned)	R, Python, TIME Also includes support for ArcGIS and related spatial functions	Fortran	Fortran 77, Fortran 90 and C	?
	Invasiveness How much a model needs to be altered before it can be used in the framework (Jagers, 2010; Lloyd et al., 2011)	Both OpenMI and CSDMS use similar methods to prepare components for use in each framework, namely implement methods such as initialise, run, describe and finalise. It was not clear from this initial investigation if one was much more invasive than the other.				?	High, this framework is designed for a set of fixed models representing the key earth systems	Low- intrusiveness, portability and flexibility are key design concepts	Low
	Time stepping	Yes	Yes	Yes	Yes	?	Yes	Yes	Yes
	Two way model communication	?	?	Yes	Yes	? Most descriptions involve linear one direction workflows	Yes	Yes	Yes
	“Non-temporal data source” e.g. 3D model files or database	Yes 68 datasets available on the CSDMS portal 25/10/2013	Yes	No	Yes	Yes	Yes Typically two dimensional gridded datasets are passed	Yes	Yes
	Model metadata The framework supports the capture of metadata, ideally at least partially automated	Yes Via a model metadata file XML		Yes OMI XML file defines exchange items, more descriptive information can also be captured		?	Yes	Yes	Yes
	Qualitative model exchange items	?	?	No	Yes	?	? None identified	? None identified	? None identified
Utilities	Spatial conversion	Yes Grid based		Yes		Yes	Yes Grid based	Yes Grid based	Yes
	Temporal scale conversion	Yes		Yes			?	Yes	Yes
	Unit conversions	Yes	Yes	Yes	Yes	Yes	No, models need to use	Via external libraries?	Yes
							standard units
	Semantic model attribution	?	Yes	?	?	?	No	?	?
	Scientific performance1	?	?	?	?	?	?	?	?
	Implementation ready? (Eg. SDK available)	Yes	No?	Yes	Yes	Yes2	Yes although limited scope for the work we undertake	Yes	No

¹There is a danger that we confuse the evaluation of the technology and the scientific robustness of the solution, especially when the solution is relatively new or designed for another purpose.

²Project Trident is a now open source project, originally set up by Microsoft, it is described as ‘a scientific workflow workbench’. The most readily available information on an implementation of the Trident software came from publications and website for the ‘Hydrologists Workbench’, an implementation developed by CSIRO, Australia. The Hydrologists Workbench was used as a proxy for the Trident software when carrying out feature analysis for the matrix, it is therefore possible that some features identified are not fully developed in the original version of the Trident code available via CodePlex.