Open andrewgettelman opened 3 years ago
The supercooled liquid from CLUBB has been a tricky one. This was supposed to be fixed by freezing in the CLUBB interface. If you take a look at:
https://github.com/ESCOMP/CAM/blob/cam6_3_012/src/physics/cam/clubb_intr.F90
Search on 'ice_macro_tend' : this is supposed to do the freezing you are talking about. But it's not freeing liquid, just removing ice supersaturation. Which it doesn't seem to be doing.
This routine could be extended to freeze supercooled liquid for T<-40˚C. That might be the best solution.
From @zmcgraw Zachary McGraw (zachary.mcgraw@columbia.edu)
I came across the issue when I made a test run with Bergeron process efficiency set to zero and found this made clouds blow up to ~2x the total SW & LW CREs then found this persisted even rerunning with Bergeron off solely in cirrus clouds (< -40C), where it shouldn’t be occurring from the start. This seems to be due to a flawed fit between the new CLUBB scheme and MG microphysics.
I did some tests a few months ago and came to the following explanation: i) CLUBB generates cloud liquid at cirrus temperatures but does not convert it to ice, leaving this to the microphysics. ii) The microphysics codes would be expected to freeze this liquid homogeneously where <-40C, but since Bergeron process precedes it in the code and is setup to be strong at these temperatures rather than non-existent, the Bergeron process removes the large majority of liquid mass <-40C instead. iii) Since Bergeron process generates no new ice crystals (as with homog. freezing), the large majority of liquid mass <-40C gets dumped onto ice crystals already in the grid box, making them larger and fall out faster. Presumably only where Bergeron efficiency is heavily weakened or ice crystals don’t already exist are new ice crystals generated instead. Hence, the fate of this mass is controlled by a bizarre balance between Bergeron and homogeneous freezing processes operating at <-40C. iv) Any attempts to turn off (or weaken) the Bergeron process at cirrus temperatures makes sufficiently cold liquid from CLUBB homogeneously freeze instead, adding new ice crystals rather than making existing ones fall quickly. The model was not tuned for this and cirrus become extremely thick and longer lived.
And some implications: a) Model tuning works specifically for the case where ice crystals in stratiform cirrus grow heavy from external liquid mass, which may generate realistic enough cirrus but without much physical basis. b) Cirrus respond to Bergeron process efficiency as it varies, which may make cirrus responsive to stimuli in unrealistic ways. c) Altering the Bergeron efficiency from the namelist ('micro_mg_berg_eff_factor') or similar methods undesirably changes cirrus clouds in addition to mixed-phase. This becomes substantial for low efficiencies (to fix only this: implement namelist efficiency multiplier to only act > -40C).
On this we haven’t come up with a full solution, though I've switched to specifying Bergeron efficiency modifications for >-40C only for mixed-phase cloud microphysics experiments. A key question seems to be what the microphysics should be doing with supercooled liquid from CLUBB. Please let me know if you have any thoughts or questions on this.
From @andrewgettelman
Thanks for the analysis. I think CAM6 was supposed to be set up as you suggest: CLUBB generates liquid, and then the microphysics freezes everything below -40, does ice nucleation and runs a WBF calculation on everything warmer than that. It should probably happen in that order, but the microsphysics is process split. So depending on the output you might be seeing something different. There is an ice_macro_tend subroutine (clubb_intr), but this does not freeze liquid, just removes excess ice supersaturation.
The current microphysics is 'micro_mg3_0.F90' in our new PUMAS (Parameterization of Unified Microphysics Across Scales) repository:
https://github.com/ESCOMP/PUMAS/tree/pumas_cam-release_v1.16
Is the version currently in CAM if you wanted to take a look (the CAM interface is in the CAM repository).
For some reason homogeneous freezing of droplets is occurring last. I'm not sure why it doesn't happen at the top with rain and snow. I'll talk with Hugh about that.
I'm not sure that switching on and off particular effects gives a true picture of the feedbacks, since there are lots of compensating effects that occur. It's fine in one sense as a model sensitivity test, but because the model may not be doing the balance of processes right, it may not be the whole picture.
Reply from Zachary:
To comment on the proposed fix of changing the order of processes: currently there's almost no homogeneous freezing in the model as very little liquid makes it past the Bergeron process. If you move homogeneous freezing to the top as is, cirrus will become vastly thicker and longer lasting, so you'd need to raise the size of frozen crystals to at least double (or an equivalent change). I think it would be preferable to generate ice crystals at the end of CLUBB itself, as it seems messy to have cloud liquid <-40C floating around between model routines (and hence this went undetected and was difficult to attribute), though either location should fix it equivalently so long as nothing is or will be going on between CLUBB and MG08.
Hi Zachary,
We do have freezing before CLUBB to eliminate supersaturation, but this is not working on new CLUBB condensate. (This is the ice_macro subroutine in clubb_intr.F90: https://github.com/ESCOMP/CAM/blob/cam6_3_022/src/physics/cam/clubb_intr.F90, line 3511). So we are getting CLUBB formed liquid at cold temperatures.
You are correct this could be fixed at the top of the microphysics, or after CLUBB. Not sure my preference, probably in the microphysics to control it. Could also be done in both places and the second call would not do anything.
Analysis:
Why do we do homogeneous freezing of droplets fo4 T<-40C towards the end of the microphysics?
See the current version:
https://github.com/ESCOMP/PUMAS/blob/pumas_cam-release_v1.16/micro_mg3_0.F90
Line 3284.
I thought it was being done (and should be done) at the beginning, when we do the precip processes (freezing, melting). (L1577).
The issue here is we have the WBF and other processes acting on liquid below -40C created by CLUBB.
I thought we also had something to control this in the macrophysics (so we should not get any liquid below -40 into the microphysics), but I cannot find that code right now.
From Hugh Morrison:
Hi Andrew, yes we do it at the end as a "clean-up step" to ensure no liquid water below -40 C after the microphysics. Arguably we should probably be doing it both before and after, though I think in practice it probably won't matter much if liquid < -40 C is converted to ice via Bergeron or heterogeneous freezing earlier or homogeneous freezing later. Of course, this would affect any process analysis...
At any rate, I'd be fine with adding homogeneous freezing at the beginning of the micro as well.
Cheers, Hugh