Phase 2: Stabilizing long runs

[LenkaNovak]

Current problems

• our long runs with dynamic surface suffer from an unphysical thermal structure and thus global circulation
• limited baroclinic instability, the atmosphere is too stable, most likely too diffusive in mid-high latitude troposphere
• overly strong lapse rate near the surface (insufficiently diffusive)
• build: https://buildkite.com/clima/climacoupler-longruns/builds/312#018b763d-ff2a-4170-b45b-87424c52558f

ClimaAtmos specs / debug

• diffusivity (week Nov 6)
  • fix long runs in Atmos [@Sbozzolo]
  • exp decrease with height, from sfc (build) and above 950hPa (build) [@LenkaNovak ] (see below)
  • Frierson parameterization in FMS dependent on Ri #509
  • Relaxation scheme for more efficient vertical transport (Betts-Miller) #510
  • Investigate non-conservation in ClimaAtmos + implement checks in CI Investigate conservation issues in aquaplanet runs ClimaAtmos.jl#2458

ClimaCoupler specs / debug

• equilibrated state with slab ocean
  • in Coupler run slab ocean for 900 days for the current target build  #513
  • check energy surface fluxes balances against observations
  • add slab ocean to ClimaAtmos longruns #511
  • check negative evaporation issue  #512
  • comparisons with the ZG23 paper #514
• versioning / updating (once physical issues resolved)
  • update to ClimaAtmos 0.16.3 [@juliasloan25] use new package versions #502
  • fix conservation after dependency updates #556 (#Small fixes for conservation #568)
    • following params of this run
  • Target EDMF-less longruns #515

Arising issues to fix

• proposed fix for LSM negative W and sigmaS #472
• race condition in land albedo read #470
• Fix discontinuous water runoff #484
• Adjust surface roughness for less abrupt initialization #604

[akshaysridhar]

Build 312: Standalone atmos run (clearsky, monin-obukhov surface scheme, time varying insolation) Averaged over days (100-300) at 10-day output frequency.

[akshaysridhar]

Build 312: Coupled atmos run (clearsky, monin-obukhov surface scheme, time varying insolation) Averaged over days (100-200) at 5-day output frequency.

[szy21]

What is the difference in the standalone and coupled atmosphere run, and what is the albedo in these simulations?

[LenkaNovak]

standa

What is the difference in the standalone and coupled atmosphere run, and what is the albedo in these simulations?

standalone = ClimaAtmos does the stepping and flux calculation (no coupler)
coupler = full AMIP (except topography and EDMF)

Albedo for standalone is 0.38 I believe, and AMIP reads this from a file. We get similar results for idealized albedo as a function of latitude.

[LenkaNovak]

Standalone only

Here are results from the build with exponential decrease of CE from the sfc: build #1231) (similar results for runs where C_E is constant to 950hPa, then exponentially decreasing)
NB: C_E = CE_0 * exp(-((1e5 - p)/1e4)^2)

day 100 zonal-mean snapshot (similar for day 180):

data:

• /central/scratch/esm/slurm-buildkite/climaatmos-longruns/1231/climaatmos-longruns/tvinsolation_diff1/day100.0.hdf5
• /central/scratch/esm/slurm-buildkite/climaatmos-longruns/1231/climaatmos-longruns/tvinsolation_diff4/day100.0.hdf5

@akshaysridhar , would it be possible to output the above diagnostics for these please?

Next suggested steps:

• try coupled runs with higher diffusivity (+ more granular analysis of sfc flux balance, why is the lw_up radiative flux so strong over the dynamic surface?) NB: diff of ~5m2/s is comparable to the axisymmetric experiments in Walker and Schneider 05
• implement the FMS diffusivity that's dependent on Ri (less diffusion at poles, more diffusion at low latitudes) - as in Frierson et al 06
• if still needed, try BM relaxation scheme

[akshaysridhar]

@LenkaNovak See https://github.com/CliMA/ClimaAtmos.jl/tree/as/frierson-diff for the updates w.r.t Frierson et al. (2006) Ri dependent boundary layer parameterization.I'm having some local test run issues but I expect to be able to resolve these soon.

[szy21]

This very old branch has an implementation of Betts-Miller that is ~85% done, in case you need it.

[LenkaNovak]

Perfect, thank you!

[akshaysridhar]

Standalone only

Here are results from the build with exponential decrease of CE from the sfc: build #1231) (similar results for runs where C_E is constant to 950hPa, then exponentially decreasing) NB: C_E = CE_0 * exp(-((1e5 - p)/1e4)^2)

day 100 zonal-mean snapshot (similar for day 180):

data:

* `/central/scratch/esm/slurm-buildkite/climaatmos-longruns/1231/climaatmos-longruns/tvinsolation_diff1/day100.0.hdf5`

* `/central/scratch/esm/slurm-buildkite/climaatmos-longruns/1231/climaatmos-longruns/tvinsolation_diff4/day100.0.hdf5`

@akshaysridhar , would it be possible to output the above diagnostics for these please?

Next suggested steps:

* try coupled runs with higher diffusivity (+ more granular analysis of sfc flux balance, why is the lw_up radiative flux so strong over the dynamic surface?) NB: diff of ~5m2/s is comparable to the axisymmetric experiments in [Walker and Schneider 05](https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2004GL022304)

* implement the FMS diffusivity that's dependent on Ri (less diffusion at poles, more diffusion at low latitudes) - as in [Frierson et al 06](https://journals.ametsoc.org/view/journals/atsc/63/10/jas3753.1.xml)

* if still needed, try BM relaxation scheme

Diff 1 (C_E=1) Directory (Build 1231) (Day 180 avg)

Diff 4 (C_E=4) Directory (Build 1231) (Day 180 avg)

Diff 4 (C_E=4) Directory (Build 1231) (Day 200-280 avg)

[szy21]

This recent paper uses an idealized model with RRTMG, no convection scheme, Frierson vertical diffusion, and slab ocean. It would be a good reference for us to compare. There are differences in details about insolation and greenhouse gases between our setup and theirs. Not sure if they are important, but even if they are, it should be straightforward to modify our setup. I'll check if I can get the data from the model.

[LenkaNovak]

Constant diffusivity coupled runs (180d, start date 1 March):

• AMIP

  • crashed after day 20 build

• Slab ocean

  • large seasonal variations, unrealistic circulations (constant albedo), lagged thermal response (sfc T)
    

• fixed SST, bucket

  • more realistic bucket
    

• Diags:

Next (+ summary of the above)

• assess the long-term stability of the slab ocean build
• run u-dependent diffusivity but 4x the value builds
• implement the FMS diffusivity that's dependent on Ri (less diffusion at poles, more diffusion at low latitudes) - as in Frierson et al 06 PR
• implement BM scheme branch

[LenkaNovak]

AMIP with increased (u-dependent) diffusivity

• K_E x 4 and K_E x 2 yielded similar results (more extreme for the former)
• evolving surface too cold (and dry) at the poles and too warm (and wet) at the equator
• lapse rate excessive in the tropics, very stable at the poles up to 1km (though not unrealistic)
• both crash at day 120 (peak summer)

[LenkaNovak]

Slab ocean with increased (u-dependent) diffusivity

• K_E x 4 crashed at 40d and K_E x 2 did not crash (ran to 320d)
• higher diffusivity causes the high latitude temperature to be too warm, jets and eddies are more distorted / banded and also much stronger (less extreme lapse rate), superrotation

Current status

• higher diffusivity makes the tropic less unstable but the poles less stable. This points to the need for the Ri-dependent diffusivity, and potentially more extreme transport in the tropics
• we also checked the surface fluxes (rad and turbulent) against observations and these seem to be reasonable, except that the latent heat release is too high over land (max: 150W/m2 v 75W/m2) and potentially too low over the ocean (max: 100W/m2 v 150W/m2)

Next

• implement the FMS diffusivity that's dependent on Ri (less diffusion at poles, more diffusion at low latitudes) - as in Frierson et al 06 ClimaAtmos PR
• reproduce existing diffusion only GCMs (e.g. Zurita-Gotor et al 23 - see above)
• examine negative evaporation in coarse resol runs
• BM relaxation scheme