Session 3 Breakout 2 : What could we get out of CMIP7

This topic contains discussions, questions and thoughts for Session 3 - Breakout room 2 of the CMIP7 Workshop
Time and Date: Tuesday 28th February, 2023 from 2:00 - 3:30 pm

CC: talks this morning strongly supportive of cmip7 participation - but why?

HC: devils advocate - what benefitto Aus for us running models rather than downloading output from elsewhere?
CHS: hi-res regional projections expensive, esp BARPA - will large scale signals be improved in new global models?

AU: local benefits in running Aus-specific dynamics eg. in land surface; also in having expertise in using and critiquing model output

AU: model documentation is incomplete - need local expertise from actually running models to be able to understand

HC: is that an argument to be involved in CMIP7 or to instead build the best model for Australia?

AU: still important to benchmark against other models

HC: max scientific benefit obtained if documentation is properly done

CM: very challenging to get the esdoc pulled together - CMIP process broke down; hard to get all the needed contributions - all done voluntarily - resourcing documentation is an issue, not something you can tack on at the end

HC: we can feed back our recommendations to CMIP; and this meeting is an opportunity to put resourcing in place ahead of time

CHS: if we remain involved our feedback will be more convincing

AK: we really need to keep the capability for climate modelling development as climate expertise will be crucial for years to come.

AK: We need to have capability in Aus for these models, to train people up and evaluate model output. Climate challenge goes over multiple generations - need to have cohorts that come through that know how to deal with these systems.

CHS: impact not only for climate but all the systems that depend on this

MT: capability to run models ourselves - can’t expect anyone else to run models for us if we need different output etc

RT: model spatial resolution - tradeoff between resolution and ensemble size; higher resolution is crucial for local planning etc - would a (say) 50km global model be better for driving downscaling?

CM: data volumes: even at low resolution ACCESS generated 2PB of raw output, but ~500Tb on ESGS; any resolution increase becomes prohibitive, esp. with the number of variables and ensemble members required. Many tradeoffs.

CC: ESM1.5 a popular download due to large ensemble, which is because it was cheap to run

CHS: we haven’t had a step change in data storage for CMIP6 so expect the same issue for CMIP7

HC: we are limited in the combination of resolution, ensemble size, run length, variables, etc

CC: developing a high res model as well as nominal 1deg is a big task - new tuning etc etc

HC: developing high res could be a challenge

CHS: machine learning, new statistical methods, etc - will NCI capacity compute & storage be a limit? What gaps are the new techniques needed to fill?

RT: yes, tradeoffs between various objectives, but progress on resolution has been slower than it could have been and is yet to be fine enough to be directly useful for local planning etc, and might also be undermining the accuracy of downscaling

CM: need to discuss this tradeoff - eg is one high res more useful than 50 ensemble members?

HC: need to find and optimum/compromise

CHS: each community has different requirements

HC: There is benefit of running more high-resolution runs if it can match our HPC requirements.

HC: consider the counterfactual: what would we lose if we didn’t do CMIP7?

RT: Aus models are dryer than average - widens the envelope - important to contribute this scenario so we consider this possiblilty

CHS: agreed

RT: we appear at the very dry end in CMIP comparisons and we struggle with drought - so this could be especially important for us

HC: having this spread is important for quantifying risks to aus