...
Understanding The ACME model:
...
Tier 1b diagnostics (Xie_ACME_Atmos_Diagnostics.pptx)
- During tuning, we used AMWG for diagnostics. Now that tuning is done, we should do more diagnostics and analysis of the model when v1 is complete. We should revisit the Tier 1a, Tier 1b diagnostics and produce a package that everyone can run, not just the PIs who wrote the code.
- Tier 1a is done.
- Reviewing the Tier 1b diagnostic collections:
- Some PIs left. (Jin-Ho Yoon, Tianyu Jiang)
- Some are listed but don't have anyone working on them
- If someone is listed as working on a diagnostic collection but doesn't have time to do it, please let team leads know.
- Phil: we originally identified these as particularly important for ACME science and planned to use them for tuning the model. If we get these right, then the next time we build a model, we will optimize for these things. Right now, we're still doing the tuning using the old strategies and not using the new approaches. But there is no way to use the new approach until these things work out-of-the-box. Once these work, we could use these to develop objective metrics for model tuning.
- Shaocheng: we did use some of this (e.g. Tier 1a) when doing the convection assessment. We just need a little effort to get these things done. Once v1 tuning is complete, you can expect that Phil and I will push you to really get them done.
- Scott: Wouldn't the idea be to build these tests to be run automatically, similarly to unit tests?
- Phil: the unit tests are checking to be sure that the software is operating correctly. This exercise is to tell us about the fidelity of the simulations. IN the ideal world, we would have a workflow that not only runs the model, but runs the diagnostics in a fashion that is as automated as possible.
- Scott: in addition to a human looking at it, building really objective diagnostics would be valuable, e.g., what is an acceptable error in precipitation over the Amazon, some overall performance score, where some statistically clever person has looked at how to diagnose what is an acceptable performance.
- Shaocheng: next step is that we need to make sure that Tier 1b diagnostics that exist already can be run by any team member, not just the PI. Peter and Kate will discuss this further.
Using the ACME model to further ACME science that does not involve the coupled simulations ( Ruby_ACME_WaterCycle_Science.pptx )
- Some science questions (e.g. aerosol indirect effect) don't require coupled simulations.
- Ruby:
- effect of high resolution on simulations