Latent heat bug in COARE atm/ocn flux routine for CESM

[nusbaume]

In the CESM COARE atmosphere-ocean flux calculations, the surface latent heat flux is calculated like so:

hlb=-rhoa * Le * usr * qsr

With the Le variable representing the latent heat of vaporization, and the rest of the terms combining to be the surface evaporation (water vapor mass) flux. In COARE Le is the "constant" latent heat of vaporization with an applied ocean surface temperature adjustment:

Le = loc_latvap - .00237e6_R8 * (ts-273.16_R8)

where loc_latvap is the constant value, and ts is the ocean surface temperature. The latent heat flux (hlb) term is then passed out from COARE, and converted back into an evaporation flux for use by CESM. However, that conversion doesn't use the the temperature-adjusted Le variable, but instead the constant value:

evap(n) = lat(n)/loc_latvap

where lat(n) is the same as hlb. This results in an Le/loc_latvap, or 1 - .00237e6_R8 * (ts-273.16_R8)/loc_latvap, bias in the evaporation flux to CESM relative to what COARE actually predicts.

The fix is either to just pass evap out directly from COARE, or simply use Le instead of loc_latvap in the lat->evap conversion.

Credit goes to Benjamin Barr for confirming the existence of the bug!

[billsacks]

@megandevlan FYI

[JensBDebernard]

I don't think this is a bug, but an intentional consequence of the underlying assumption in CESM that phase transitions are calculated relative to 0 deg C. The temperature dependent term for Le in the COARE scheme is to correct for the actual temperature of the mass evaporated, and it equals the difference in sensible heat for liquid water and water vapor at a given temperature, relative to 0 degree C. (See similar derivation in e.g. equation 15 in https://rmets.onlinelibrary.wiley.com/doi/10.1002/qj.3899).

The compromise done when using COARE is to iterate the sensible and latent heat fluxes with respect to the true ocean temperature, but then, converting the resultant latent heat flux to a vapor flux that is consistent with the overall assumptions in the coupled model in cmeps. In that way, the vapor flux, and latent heat flux is consistent, while the only inconsistency is between the vapor flux within the COARE routine (which is not used elsewhere), and the vapor flux used in the rest of the model. The latent heat flux might be different from what would be obtained by using a constant Le directly in the COARE routine, but it is not obvious that the latter approach would be better.

The NorESM simulation NorESMhub/noresm3_dev_simulations#484 tests the impact of using Le to calculate evap. This clearly gives a heating of the atmosphere relative to the control experiment, and non-conservative heat source of the order of 1 to 2 W/m^2 globally averaged. This is consistent with the 2-3 % difference in evap when using Le, opposed to loc_latvap, with typical subtropical - tropical ocean surface temperatures.

This behavior should off-course be changed when a both mass and enthalpy conserving exchange between the components is implemented.

[nusbaume]

@JensBDebernard @mvertens @oyvindseland @DirkOlivie @matsbn Thanks for running those experiments and helping uncover the core assumptions! I totally didn't realize that 0 deg C assumption existed, although it makes sense looking at some of the component code, so I apologize for the lost time and effort on your end.

Given that it seems that several people (myself included) on the CESM-side didn't catch the reason for this mass flux adjustment, I will try to add a comment in the code explaining it, so that this doesn't pop up again (at least until a larger effort is taken to ensure better conservation across all components, as mentioned). I'll leave this issue open until I get the comment brought in, at which point I'll make sure to close it.

Thanks again for your help!

[PeterHjortLauritzen]

This is the issue of different components/parameterizations making different thermodynamic assumptions: COARE uses variable latent heats with a liquid reference state and a 0°C reference temperature, MOM6 uses constant latent heats with a liquid reference state and a 0°C reference temperature, and the atmosphere uses an ice reference state and a 0 K reference temperature.

To avoid inconsistencies, it is probably easiest to pass evaporation around and let each component compute whatever latent heat flux is consistent with its own formulation (which is what we are currently doing), together with their respective fixers.

We have aspirations to move the atmosphere to variable latent heats (consistent with COARE), but we would still have to make sure to use component-consistent reference states (water phase and reference temperature).