Science challenges - streamlining model selection workshop by

Example post

cmipipo — 2025-02-02 15:28:38 UTC

You can add links, pictures etc and then write your comment or question and hit Publish!

Colin Jones

cmipipo — 2025-02-05 20:17:22 UTC

E-driven models - way CORDEX select data you likely would not select e-driven models but are more applicable to future so need to think about applicability of future as well as representation of the past

Emission scenarios

2025-02-05 20:18:05 UTC

Decsion about scenarios; might be more difficult in CMIP7

Taking a step back: Do we want a common set of criteria for model selection?

2025-02-05 20:18:26 UTC

Need for central source of derived quantities

cmipipo — 2025-02-05 20:18:31 UTC

Gavin Schmidt: How hard to assess when data not all in one place. Climate sensitivity/metrics, skills of ENSO, other derived quantities should be available somewhere.

Definition of uncertainty

cmipipo — 2025-02-05 20:20:18 UTC

Important to be clear about what we mean by uncertainty

Rapid Evaluation Framework

2025-02-05 20:21:09 UTC

Will the REF be a 'game changer' for model selection?

Selection critera

2025-02-05 20:22:09 UTC

Choosing 'families' of criteria

2025-02-05 20:23:49 UTC

Selection criteria

2025-02-05 20:24:11 UTC

difficult to find a common strategy in different activites
climate sensitivy is often used
find common set and then critera that are specific for regions
ECS might be different for region compared to global
model familiy; models can be different even if they share components
use same set of critera across activities and use specific critera for regions
be pracmmatic; data must be avail.
length of simulation might also be an issue
understand stakeholder needs is important; model selection might depend on that

Should we do model selection

cmipipo — 2025-02-05 20:24:59 UTC

By doing model selection in the first place, could be losing important future information, e.g. tipping points

Regional patterns in SST change

cmipipo — 2025-02-05 20:25:05 UTC

John Dunne: In CMIP7 hope to address representing regional SST and higher res to represent extremes.

High res = improved physics??

cmipipo — 2025-02-05 20:26:12 UTC

Alberto Meucci: High resolution is not only aspect to give the extremes but also improved physics. How much is confidence around high resolution models?

Global vs regional criteria?

2025-02-05 20:27:12 UTC

data for selecting models based on GCM processes

hayashimichiya — 2025-02-05 20:28:37 UTC

Could we create a table that summarizes the implementation of GCM model components, such as aerosols and clouds? This would help us select models that demonstrate a wide range of process differences or a more consistent level of process reliability.

Process representation

cmipipo — 2025-02-05 20:28:46 UTC

Decisions can be made on

1) whether a process is represented

2) how well the process is represented

Uncertainty quantification

cmipipo — 2025-02-05 20:29:41 UTC

Often a lack of observations which can limit you, especially when considering different geographical regions

Two levels, one generic and one region/initiative-specific for model selection

2025-02-05 20:30:57 UTC

We could think of two levels, one of common metrics/criteria from a global perspective, common for all initiatives (sensitivity, biases, etc.), and another region/initiative-specific, dealing with the specificities of each initiative (e.g. key regional drivers for CORDEX, etc.). The first level, combined with data availability could yield a sub-ensemble of common models for all initiatives. The general level could be coordinated with/aligned to the Rapid Evaluation Framework (REF):

Possible for CMIP7

2025-02-05 20:31:29 UTC

ESMValTool might be good starting point
Rapid Evaluation framework in CMIP7 will also be an option
for CORDEX it is important to know early which groups save boundary conditions for downscaling
regular updates from the groups on the status of the simulations from CMIP would be useful

ECS, model families

2025-02-05 20:32:20 UTC

ECS: regional results are rather independent of the driving models ECS; plausible range normally includes the 10-90% range but not the full range, thus maybe not a very good criteria to disregard models.

Model families: selected ESMs from Euro-CORDEX include many NEMO-models in the ocean. These show quite different results at the regional scale. Thus, having one component in common should not be a criteria to disregard models.

comparison with observations

2025-02-05 20:33:00 UTC

core scenarios across different activities

2025-02-05 20:34:18 UTC

It would be useful to decide on a set of ESMs that would be used by both CORDEX, ISIMIP .... This would help to provide consistent information to IPCC, users ....

Who should do the selection?

2025-02-05 20:34:21 UTC

meenu19 — 2025-02-05 20:35:02 UTC

As an academician and hydrologist researcher, a lot of research queries that we face are reflecting the communication of uncertainty that further propagates to applications in hydrology. So if we could borrow the existing time-tested methods, from modellers and convince ourselves the local impacts etc.

Challenges when with selecting the best criteria

2025-02-05 20:35:46 UTC

Energy Sector

cmipipo — 2025-02-05 20:35:50 UTC

How does the energy section look at the uncertainties and choose what output to use for planning?

data availability

2025-02-05 20:37:20 UTC

we need to ask us why not more ESMs provide the required boundary data? One reason is that it is extremely expensive, 10 times more data need to be stored for dynamical regional downscaling. Not choosing the ESMs that make the effort to store and cmorize all data will lead to even less ESMs that will provide the data.

Glacier models

cmipipo — 2025-02-05 20:40:20 UTC

We no longer use SSP scenarios to show results, but instead look at Global Warming Levels, and choose models which fit this. For this we need other uncertainty assessments.

Collaborating with the global community models

2025-02-05 20:40:27 UTC

cmipipo — 2025-02-05 20:47:24 UTC

Here is the paper containing the plot I mentioned (Fig 3): https://esd.copernicus.org/articles/12/1543/2021/esd-12-1543-2021.pdf

What would make the biggest difference for CMIP7?

cmipipo — 2025-02-05 20:47:35 UTC

Internal variability sampling

jesusfernandez20 — 2025-02-05 21:06:51 UTC

regarding the sampling of internal variablity mentioned by Paco, there's also a limitation on the availability of LBCs to feed dynamical downscaling: https://wcrp-cordex.github.io/cmip6-for-cordex/CMIP6_studies_table_EUR.html#all

Still, there are some efforts to downscale different members: https://wcrp-cordex.github.io/simulation-status/CORDEX_CMIP6_status_by_experiment.html#EUR-12-EURintvar

There could be room for ML/AI here

Regional Impact & Spatial Resolution

tantrayfayaz — 2025-02-06 07:32:45 UTC

Models need to be more of Regional Based and the Spatial Resolution need to be improved based on Integration of AI & ML Technologies

Emergence of coupled(ESM) regional models?

2025-02-10 03:01:47 UTC

Traditionally our regional CORDEX models have been atmospheric only, with targetted regional ocean, sea ice etc models. Is there a view to moving toward coupled regional models in CMIP7?