[CIG-MC] AGU's new data policy

[Scott King]

Hi also agree that movies can be useful, although in some cases not so interesting as one might think (strongly internally heated models can be a "nothing burger” in isotherms but you can find large-scale structure that develops looking at the geoid or dynamic topography.  I’m unfamiliar with the Afonso et al. paper so I will need to look at it.  Sounds interesting, thanks for pointing it out.  My concern (without having read it) is that sometimes in time-dependent fluid simulations, to be reproducible you need the smallest bits because that is where instabilities grow from (think chaos).   In that regard, I’m not even sure we have a good definition of what “reproducible” really means.  Are we sure that the same code on a vastly different compiler with different versions of the libraries will produce the same result?   Circa 10 years ago this was not clear with some NWS codes.  They were certified with specific compiler versions.  The computer science community is wrestling with some of these issues in regard to large simulations (i.e. what Louise calls leadership class) because there are groups that are pursuing the idea that you want an algorithm that can continue on “successfully" even if/when one node fails.  Understand that this new policy is broader than AGU.  The policy is called FAIR and stands for Finable, Accessible, Interoperable, and Reusable.    I think the ideas are excellent and worthy and we as a community should get behind them and in many cases, this is a good idea.  There is an article and links to more information here so people can actually see what the policy (which seems to be evolving) really is...

https://eos.org/agu-news/enabling-fair-data-across-the-earth-and-space-sciences


> On Mar 13, 2018, at 6:01 AM, Thomas Ruedas <Thomas.Ruedas at dlr.de> wrote:
> 
> I agree that movies are often useful for providing insights into results that may not have been included in a paper, but they cannot be more than a convenient supplement for a data archive, because they do not provide the quantitative information in a format that can be reused by others. Providing the actual datasets, by contrast, would allow others to generate movies if they want.
> A more practical solution for producing a "small" archive is probably the use of lossy compression formats. I recall a paper in G3 by Afonso et al. (2015) (http://onlinelibrary.wiley.com/doi/10.1002/2015GC006031/abstract) in which a method was presented that allows the efficient compression of multidimensional datasets. Although the paper had phase diagrams in mind, this or a similar method could probably be used for storing modeling results without any substantial modifications at a lower cost than the storage of the raw, uncompressed data. Such an archive would not store the original raw data, but it would provide a dataset that should allow the reproduction and, more importantly, the replication of the published results. The advantage of replicability over reproducibility was brought up by Louise Kellogg earlier in this thread, and I concur with her that this is what AGU should aim for. Maybe it would be good if the community could settle on a data compression standard and provide tools for it.
> 
> Best,
> Thomas
> 
> Am 13.03.18 um 09:40 schrieb Magali Billen:
>> What are people’s thoughts about using movies of simulation output as a way of archiving the model results?
>> These have to put together well to document the model results, but together with the ability to rerun a model, they are a much smaller way to archive results.  I also find that in reviewing papers that have time-dependent models, its nice to see movies that document the full dynamics  beyond what can be shown
>> in snap-shots. Often what we put in the paper depends on what we are focused on in that particular manuscript, but there is often a lot more that could later be “mined” from model results. Making the
>> movies allows others to see the data mining potential of the model results, for questions that we might
>> not have consider ourselves. This of course is more difficult for 3D simulations, but I still think it could
>> be a good way of making the model results more transparent and useful.
>> Thoughts - should movies be part of a CIG recommendation for how to archive results for manuscripts?
> 
> 
> -- 
> orcid.org/0000-0002-7739-1412
> _______________________________________________
> CIG-MC mailing list
> CIG-MC at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-mc