Version 17 (modified by Martin Dix, 6 weeks ago) (diff)


I've done a one year CABLE run starting from a Jan 1 initial state from the au-aa655 CABLE AMIP run and compared to a similar one year JULES run also starting from an AMIP initial state.

My CABLE run is basically a copy of Jhan's u-ap748 with a couple of tweaks to the reconfiguration to enable it to work with the AMIP file.

Time series of various global means

NetCDF files with the monthly data are /short/p66/mrd599/ (JULES) and (CABLE).

In the annual mean the TOA net flux is 1.2 W/m2 with JULES and -2.3 with CABLE, consistent with the cooling.

Net heat flux in bare soil regions is about 5 W/m2 in CABLE, even though the depth averaged soil temperature doesn't change much over the year.

To further investigate the surface energy balance I set up JULES and CABLE versions of the single column model, suites u-ar806 and u-ar807 respectively.

Used a configuration with bare soil, zero initial soil moisture and relaxation to initial atmospheric state to allow long runs with an extreme diurnal cycle.

In routine stempv, added a calculation of the integrated soil heat content and the initial and corrected ground heat flux.

   ! Integrated heat content
   tmp = 0.
   DO k = 1, ms
      tmp(:) = tmp(:) + ssnow%tgg(:,k)*ssnow%gammzz(:,k)
   write(6,*) "COLUMN HEAT", tmp
   write(6,*) "DT", ssnow%tgg(:,1) - tgg_save(:)
   write(6,*) "GA", canopy%ga, canopy%dgdtg, canopy%ga + canopy%dgdtg*(ssnow%tgg(:,1) - tgg_save(:))

This can be compared with the net surface heat flux calculated from the SCM diagnostics surfsw + surf_lw - sens_ht - lat_ht.

The old access 1.3 version of the SCM doesn't have quite the same options for atmospheric relaxation but it's possible to set up a case that has a similarly large diurnal cycle and soil warming.

Averaged over the last 40 days of the runs

JULES 10.6 CABLE 10.6 CABLE 7.3
SCM net 4.41 10.87 6.77
Soil change 4.36 5.18 6.51
GA 5.87 5.69
GA corrected 5.22 6.51

In access 1.3 there was a good agreement between the SCM net flux and the soil temperature change with the higher frequency radiation making almost no difference. In 10.6 CABLE seems to be losing about 5 W/m2.

The UM has a net radiation on tiles diagnostic that was not saved by the SCM for some reason so I added it. In both JULES and CABLE this uses the updated surface temperature to calculate the upward LW, so should better match the soil temperature change. CABLE 7.3 results from here on are using hourly radiation.

Grid box net radiation 95.14 73.19 33.60
Tile net radiation 95.12 67.45 33.26
Tile net flux 4.40 5.13 6.47
Soil change 4.36 5.18 6.55

Therefore it seems that using the radnet_tile diagnostic the energy balance is equally good in JULES and CABLE. However this isn't the flux that the atmosphere uses. Neither JULES nor CABLE alter the grid box mean LW which is what affects the atmosphere.

Comparing the fluxes is simpler if the radiation is called every time step. This has only a small effect on results and doesn't change the CABLE 10.6 imbalance.

Grid box net radiation 95.44 73.24 33.78
Tile net radiation 95.13 67.47 33.43
Grid box net flux 4.69 10.91 6.84
Tile net flux 4.39 5.14 6.49
Soil change 4.39 5.23 6.58

Following plots are the mean diurnal cycle over the last 40 days of the runs

Checking CABLE fluxes in more detail pointed to inconsistencies in the emissivity. cable_data.F09 has

     emleaf = 1.0, & ! leaf emissivity
     emsoil = 1.0, & ! soil emissivity

but rad_ctl.F90 has code that uses the input emissivities

       ! Types match tiles here.
        emis_tiles(1:npft) = emis_pft
        emis_tiles(npft+1:ntype) = emis_nvg

At the moment these are still set to the JULES values in the model configuration with only a change for the number of PFTs.


Setting all these to 1 in the SCM tests gives a much better energy balance.

CABLE emiss corrected
Grid box net radiation 67.63
Tile net radiation 67.36
Grid box net flux 5.42
Tile net flux 5.14
Soil change 5.23

There's now almost no difference between the CABLE radnet_tile and the grid box net radiation, showing that CABLE isn't updating it for the change in surface temperature, though it used to back in access 1.3. However as far as I know this only affects diagnostics.

Note that this emissivity problem should have also affected the vn8.5 coupled model runs with CABLE because the namelist for that had the same JULES emissivities.

The immediate effect of change is to increase the upward LW at the surface (and at the TOA), restoring the surface energy balance by reducing the diagnosed net radiation. However this doesn't immediately change the radiation flux into the surface because CABLE was already calculating that using the correct emissivity. Surface changes should only occur as a response to atmospheric warming from the increased upwards LW flux.

u-as614 is a rerun of u-ao854 with the emissivity fixed. There's very little difference to the surface temperature. However the net surface energy flux averaged over the year is generally smaller.

Comparison with vn8.5

There was a suggestion that the coupled model with CABLE in vn8.5 (GA6 rather than GA7.1 atmospheric configuration) didn't show the same global cooling so I set up SCM tests with vn8.5 as well. UMUI build jobs are vaams (JULES) and vadvs (CABLE).

Results here from 50 day runs with atmospheric relaxation and the GABLS3 surface initialisation (C3 grass with LAI=2). These runs have almost no precipitation so the soil steadily dries.

The CABLE surface temperature and latent heat flux show several jumps, probably when the soil moisture in various layers reaches the wilting point.

Following figures show the mean diurnal cycle over the last 20 days when things are reasonably smooth.

Layer soil moisture changes

Attachments (27)