Radiation balance

[MartinDix]

Latest CM3 runs (e.g. cm3-run-16-09-2024) have a net TOA energy flux around 2.7 W/m^2 initially.

Met Office GC5 AMIP runs were tuned to give the desired radiation balance at N216, not N96. The tuning is described at https://code.metoffice.gov.uk/trac/gmed/wiki/ticket/573/TicketDetails describes the radiation tuning of the N216 version.

This is an extract of the table near the end of https://code.metoffice.gov.uk/trac/gmed/wiki/ticket/577/TicketSensitivityTests

	rsut	rlut	net
Obs CERES EBAF 4.1	99.0	240.3	+1.0
N96 RC3	98.56	239.53	+2.31
N216 RC3	99.07	240.59	+0.74

RC3 is almost the same as the final version, except for an update to the boundary layer buoyancy https://code.metoffice.gov.uk/trac/jules/ticket/1242.

Alejandro Bodas's presentation at the 2023 Momentum Workshop says they retuned the GC5-N96ORCA1 PI control aiming at rsut=99, rlut=240.5, close to CERES EBAF4 values (because UKESM suggested PI and current were pretty close). Used two_d_fsd_factor = 1.70, cape_timescale =2700 s in place of standard two_d_fsd_factor=1.65 cape_timescale=1800.0, though not clear if this is the final value.

[MartinDix]

The Met Office n96orca1 PI control, e.g. u-ct598, instead tunes some cloud ice detrainment parameters. Compared to the AMIP configuration we've used until now changed parameters are

Parameter	AMIP	PI control
cff_spread	1.0e-5	1.0e-10
tice	263.15	253.15
starticeTKelvin	263.15	265.15
alliceTdeg	-20	-32
two_d_fsd_factor	1.65	1.67

From the UM metadata

• cff_spread_rate: Ice cloud fraction spreading rate
  When the effects of shear on falling ice cloud are either ignored or derived from model winds, this controls the rate at which ice cloud fraction spreads.

• starticetkelvin: Start detraining condensate as ice at (Kelvin)
  When the method for coupling PC2 to convection scheme is 3 or later, this sets the temperature at which detraining condensate as ice begins.

• allicetdegc: All condensate is detrained as ice by (deg C)
  When the method for coupling PC2 to convection scheme is 3 or later, this sets the temperature at which all condensate has detrained to ice.

• tice: Specify phase change temperature /K
  Use a phase change other than 0 degrees - choose this if running with PC2. The value commonly used with PC2 is 263.15.

• two_d_fsd_factor: Factor to convert 1D FSD for straight line to 2D value
  Current parametrizations of FSD (fractional standard deviation, a measure of subgrid variability) of cloud water content used in radiation and microphysics schemes are based on observations taken in a 1D line. For the UM we require the FSD for a 2D gridbox, which is generally larger. We account for this difference by multiplying the 1D fsd by this value.

[kieranricardo]

Run cm3-run-29-10-2024-radiation-balance with two_d_fsd_factor = 1.70, cape_timescale =2700 didn't show any improvement.

Run cm3-run-06-11-2024-radiation-balance-ice-detrainment with ice detrainment started.

[kieranricardo]

TOA flux not looking promising over 5 years. I've extended the run, and it's currently running for 20 years.

[MartinDix]

Changes in the right direction, but only a small improvement from these parameters in an AMIP run.

	Std	New options	CERES
rsut	98.41	98.84	98.93
rlut	239.62	239.82	240.35
net	2.37	1.74

[jemmajeffree]

I updated the current CM3 100km resolution with all of the parameters in rose-app.conf that were explicitly changed in the GC5 configuration (See Martin's post here #53 (comment)). I ignored any changes outside rose-app.conf, and any parameters that were added, which left me with a hack-job piece of code I'm including here for reproducability:

Code to change rose-app.conf file

   replacement_pairs = ( # Manually created list of every replaced variable
('f0_io=0.875,0.875,0.900,0.800,0.900','f0_io=0.78668,0.78668,0.80916,0.71925,0.80916'),
('nl0_io=0.040,0.030,0.060,0.030,0.030','nl0_io=0.0605,0.0453,0.0907,0.0453,0.0453'),
('rootd_ft_io=3.00000,1.00000,5.00000e-1,5.00000e-1,5.00000e-1','rootd_ft_io=3.0,1.0,0.5,0.5,0.5'),
('tupp_io=36.0,31.0,36.0,45.0,36.0','tupp_io=31.328,26.977,31.328,39.160,31.328'),
('z0v_io=1.1,1.1,0.22,0.22,1.0','z0v_io=1.10,1.10,0.22,0.22,1.00'),
('a_snow_et=2.8e-6','a_snow_et=2.80E-06'),
("!!rp2_seed='unset'","rp2_seed='unset'"),
    ('cbl_op=2', 'cbl_mix_fac_nml=0.0\ncbl_op=4\ndec_thres_cloud=0.090357849'),
    ('dbsdtbs_turb_0=1.50e-4','dbsdtbs_turb_0=0.000625274'),
    ('cape_timescale=1800.0','cape_timescale=2090.08049'),
    ('cpress_term=0.3','cpress_term=0.930400342'),
    ('efrac=1.0','efrac=0.890258336'),
    ('mparwtr=1.0000e-3','mparwtr=0.001074432'),
    ('prog_ent_grad=-1.1\nprog_ent_int=-2.9\nprog_ent_max=2.5\nprog_ent_min=0.5\nqlmin=4.0000e-4',
     'prog_ent_grad=-0.57728338\nprog_ent_int=-1.197885628\nprog_ent_max=1.636050439\nprog_ent_min=0.698672135\nqlmin=0.000278709'),
    ('r_det=0.5000','r_det=0.645433628'),
    ('thpixs_mid=0.5','thpixs_mid=0.762050889'),
    ('fbcd=4.00\ngsharp=0.70\ngwd_frc=4.00','fbcd=2.087563327\ngsharp=1.67236774\ngwd_frc=4.537798114'),
    ('nsigma=2.6','nsigma=3.138604429'),
    ('ai=2.5700e-02','ai=0.019311345'),
    ('aut_qc=2.47','aut_qc=3.22612713'),
    ('c_r_correl=0.9','c_r_correl=0.433247206'),
    ('aparam=0.077','aparam=0.090079778'),
    ('bparam=-0.1365','bparam=-0.138289901'),
    ('dp_corr_strat=10000.000','dp_corr_strat=35851.17136'),
    ('two_d_fsd_factor=1.65','two_d_fsd_factor=1.268464718'),
    ('stag_dropoff=2.0','stag_dropoff=1.221073607'),
    ('!!a_ent_1_rp=0,0,0','a_ent_1_rp=0.470377275,0.470377275,0.470377275'),
    ('!!charnock_max=0','charnock_max=0.018'),
    ('!!charnock_min=0','charnock_min=0.018'),
    ('!!g0_rp=0,0,0','g0_rp=33.01247062,33.01247062,33.01247062'),
    ('!!g1_rp=0,0,0','g1_rp=1.738350798,1.738350798,1.738350798'),
    ('!!i_rp_scheme=0','i_rp_scheme=0'),
    ('l_rp2=.false.','l_rp2=.true.'),
    ('!!l_rp2_cycle_in=.false.','l_rp2_cycle_in=.false.'),
    ('!l_rp2_cycle_out=.false.','l_rp2_cycle_out=.false.'),
    ('!!lambda_min_rp=0,0,0','lambda_min_rp=42.29578972,42.29578972,42.29578972'),
    ('!!m_ci_rp=0,0,0','m_ci_rp=1.393,1.393,1.393'),
    ('!!par_mezcla_rp=0,0,0','par_mezcla_rp=0.158609211,0.158609211,0.158609211'),
    ('!!ran_max=0\n!!rhcrit_max=0\n!!rhcrit_min=0\n!!rhcrit_ref_level=3\n!!ricrit_rp=0,0,0\n!!rp2_callfreq=10800',
    'ran_max=1\nrhcrit_max=0.916\nrhcrit_min=0.916\nrhcrit_ref_level=3\nricrit_rp=0.302915832,0.302915832,0.302915832\nrp2_callfreq=10800'),
    ('!!rp2_decorr_ts=216000.0','rp2_decorr_ts=216000.0'),
    ('l_ukca_scale_marine_pom_ems=.false.','l_ukca_scale_marine_pom_ems=.true.'),
     ('l_ukca_scale_sea_salt_ems=.false.','l_ukca_scale_sea_salt_ems=.true.'),
    ('!!marine_pom_ems_scaling=1.0','marine_pom_ems_scaling=1.4'),
    ('!!sea_salt_ems_scaling=1.0','sea_salt_ems_scaling=1.4'),
)
    
with open('/home/156/jj8842/CM3-initial_testing_stuff/access3-configs/app/um/old-rose-app.conf','r') as file:
    baseline = file.read()
new = copy.copy(baseline)
for p in replacement_pairs:
    if baseline.count(p[0])==0:
        print(p[0]+' missing')
    elif baseline.count(p[0])==1:
        print(':D')
        new = new.replace(p[0],p[1])
    else:
        print(p[0]+' duplicate')    
    
with open('/home/156/jj8842/CM3-initial_testing_stuff/access3-configs/app/um/rose-app.conf','w') as file:
    file.write(new)
    

This run, which I'm calling "GC5ish" because I didn't capture all the changes, improves the 100km TOA radiation balance, though not enough to match observations. Weirdly, the 0.25° ocean is much better, with (I presume) the same atmosphere. Suggests that maybe ocean heat transport has an effect? I'd be interested in looking into where the TOA radiation balance deviates from expected, not just the global mean.

^^12 month rolling mean, as per Kieran's plot above

If anybody wants to play with the data, it's here:
/scratch/tm70/jj8842/cylc-run/ (first-pass-GC5-update is the GC5-ish updated UM, and control_025_4 is the quarter degree control)

With a bunch of non-UM diagnostics from the quarter degree moved over here temporarily to preserve scratch space on tm70 /scratch/x77/jj8842/temp_relocation_of_025_control/

[jemmajeffree]

Unfortunately, seems to get worse towards the end of the decade

[jemmajeffree]

Looking at the distribution of TOA radiation biases, I'm concerned that the CM3 atmospheric overturning circulation is much weaker than it should be.

I looked at the spatial pattern of biases in CM3 at the top of the atmosphere (rsdt = downwelling shortwave, rsut = upwelling shortwave, rlut = upwelling longwave). From this figure, I conclude that we have some huge cancelling errors in both spatial pattern and longwave/shortwave outgoing radiation. So, while the 025 and GC5-ish runs have better net TOA radiation balance, they've still got the same spatial pattern of bias of pretty similar strength.

Condensing this to a zonal mean, and adding a meridional rolling average:

From this image, I draw the following conclusions:

• There's too much outgoing longwave at the equator. Either the equator is too hot, or there's not enough/not high enough deep convection so cloud tops are too warm
• There's not enough outgoing longwave at the poles. Probably the poles are too cold. Possibly (unlikely) cloud tops are too high or too much cloud.
• Outgoing shortwave is too low in canonically "upwelling" zones — ITCZ and subpolar low — and too high in canonically downwelling zones — subtropical high.

Combined with the following image (pr = precipitation, ts = skin temperature, vas = meridional surface winds; anomalies are solid lines with the lower x axis, absolute values are dashed lines with the upper x axis):

• equator is slightly too hot in both ACCESS-CM3 default configs, but not in the GC5-ish run. Suggests to me that not enough deep convection is the main problem with the equator, supported by the fact that precipitation is underestimated around the equator.
• The poles are definitely too cold. There's also a touch too much precipitation, which isn't helping.
• Surface winds (and also MSLP, not shown here) both don't say much clearly. If anything, surface meridional pressure patterns and winds are slightly stronger than they need to be.

My conclusion is that we need more upwelling/downwelling associated with a stronger Hadley circulation. I don't yet have ideas for how to implement this. I'd like to check the Hadley circulation streamfunction, but for some reason CM3 isn't outputting 3D velocities at any frequency, so I'll need to add those to the outputs first.

[jemmajeffree]

EDIT: I should know better than to try and eyeball anomalies because I can't be bothered interpolating onto a new grid. Upwelling and downwelling may indeed be weaker in CM3. However, this comparison hinges on an approximate conversion between Pa/s and m/s and the CM3 average is only a year of data, so I'm not sure how much of what we're seeing is interannual variability in the latitude of ITCZ.
I don't think vertical velocities are the problem, they seem to line up pretty well with ERA5 in an extra 12 month run I did. (bottom right hand corner is some sort of katabatic winds off Antarctica; not relevant overall)

If anybody's running a 10 year run anytime soon, can you uncomment the following lines from app/um/rose-app.conf?

[!namelist:umstash_streq(30204_a165dc3e)]
dom_name='DP17'
isec=30
item=304
package='Standard Climate'
tim_name='T6HMONM'
use_name='UPM'

(the heaviside function needed for omega (vertical velocity in Pa/s) to be converted to netcdf) There's a few others, but that's the crucial one so I'm leaving it here for future reference.

[jemmajeffree]

I'd be interested in digging some more into spatial patterns of cloud fraction and cloud albedo, and comparing these against observations.

@MartinDix what variables would you recommend here?

[jemmajeffree]

Okay, I concede. CM3 is within CMIP6 spread for both longwave and shortwave upwelling biases. I'm slightly horrified by how consistent these biases are, and we're definitely at the extremes for the fluxes on the equator. I haven't bothered checking ESM1.5 or 1.6 here, just the models in oi10, because I'd be shocked to find it's outside the spread of other CMIP6 models and CM3.

I haven't bothered replicating here because the plot is nasty, but the upwelling and downwelling (omega; Pa/s) biases are roughly equivalent magnitude in CM3 and CMIP6 models too

[jemmajeffree]

Found this gem in the bottom of https://code.metoffice.gov.uk/trac/gmed/wiki/GCDev/GCDocumentation/GC5

To obtain a realistic net TOA for present day conditions at N96 the following changes are recommended:

• cape_timescale= 1800 -> 3600
• two_d_fsd_factor=1.65 -> 1.56

I've run 5 years; definitely seems to help but the upwards trend is concerning
Looks effective after 10 years

This is otherwise the standard 100km setup (blue); I think it's worth merging with the other tuning stuff (orange)