In the NGEE-Arctic work, and more broadly for model development in general, we have identified a need for a dashboard that allows for more complex filtering. In NGEE-Arctic, they are performing a range of simulations which are detailed here. They would need to be able to:
- Filter models by selecting/ignoring certain tasks / processes / resolutions / sites / forcings
- Filter variables by selecting tasks / processes
This addresses a problem that ILAMB visualization has--it doesn't scale well to many models. If NGEE-A were to run all combinations of things they have planned it would be 1296 different models. This is still a problem for ILAMB even in the CMIP style models. If you were to consider all ensemble members and even all models the portrait plots and dashboards become hard to parse and even view.
@jitendra-kumar and I thought through a little of how this needs to work. At first sight we might simply expand @minxu74 's unified dashboard. There may still be some UI work we can do there, but the dashboard is only a thin layer on top of ILAMB output. Adding filtering to the dashboard would have no effect on results underneath.
We brainstormed a few approaches:
- Write a
ilamb3.meta analysis that takes in a set of ilamb results along with filter options and creates a mirror of the site with only the filtered models/variables. This has the downside of a lot of replication but also a requirement for a technically savvy user. The individual who wants to see the results in a different way would need to have access to the full run and know how to execute the meta analysis with their filter options. It could work, but sounds less flexible for a science-oriented user.
- We could write a
ilamb3.meta analysis that loops over the results and builds a database (similar to REF) from which we can build visualization that populates dynamically. This would mean learning a bit of possibly React or other python-based frameworks. The downside here is that we would depend on the hosting site being able to turn on certain services to stand up the site (that is, it is no longer simple html/js).
In the NGEE-Arctic work, and more broadly for model development in general, we have identified a need for a dashboard that allows for more complex filtering. In NGEE-Arctic, they are performing a range of simulations which are detailed here. They would need to be able to:
This addresses a problem that ILAMB visualization has--it doesn't scale well to many models. If NGEE-A were to run all combinations of things they have planned it would be 1296 different models. This is still a problem for ILAMB even in the CMIP style models. If you were to consider all ensemble members and even all models the portrait plots and dashboards become hard to parse and even view.
@jitendra-kumar and I thought through a little of how this needs to work. At first sight we might simply expand @minxu74 's unified dashboard. There may still be some UI work we can do there, but the dashboard is only a thin layer on top of ILAMB output. Adding filtering to the dashboard would have no effect on results underneath.
We brainstormed a few approaches:
ilamb3.metaanalysis that takes in a set of ilamb results along with filter options and creates a mirror of the site with only the filtered models/variables. This has the downside of a lot of replication but also a requirement for a technically savvy user. The individual who wants to see the results in a different way would need to have access to the full run and know how to execute the meta analysis with their filter options. It could work, but sounds less flexible for a science-oriented user.ilamb3.metaanalysis that loops over the results and builds a database (similar to REF) from which we can build visualization that populates dynamically. This would mean learning a bit of possibly React or other python-based frameworks. The downside here is that we would depend on the hosting site being able to turn on certain services to stand up the site (that is, it is no longer simple html/js).