[Liwg-core] A possible unintended (and bad!) consequence of the P1 modification

Jeremy Fyke fykej at ae.ca
Wed Dec 5 20:50:54 MST 2018


Hi guys

Perhaps need to be careful here not to reintroduce unrealistic SMB gradients across CAM/CLM grid cells.  Eg similar to sawtoothing during previous efforts that resulted in design of accumulation/ablation specific conservation correction scheme.  My impression is that the snow-directly—to-runoff scheme could result in similar issues?

Jeremy


________________________________
From: Liwg-core <liwg-core-bounces at cgd.ucar.edu> on behalf of Bill Sacks <sacks at ucar.edu>
Sent: Wednesday, December 5, 2018 12:09 PM
To: <liwg-core at cgd.ucar.edu>
Subject: Re: [Liwg-core] A possible unintended (and bad!) consequence of the P1 modification

Hi all,

Okay, I just talked with Bill Lipscomb about this, and his understanding is that the current behavior is correct. i.e., the fact that we might have this:

For example, imagine a grid cell where CAM is generating rain at a grid cell mean near-surface temperature of 3°C; CAM's topographic height here is 1000 m. On the CLM side, we have two elevation classes, one at 0 m (50% of grid cell) and one at 2000 m (50% of grid cell). In the 2000 m elevation class, the downscaled temperature will be –3°C. This will result in all rain being converted to snow. However, with the new formulation, in the higher elevation class, rather than receiving precipitation in the form of snow, it would instead receive no precipitation.
is actually intentional and desired. (Though I should clarify that "receive no precipitation" really means, "precipitation immediately runs off".) His justification is: If CAM is generating rain at a given elevation / temperature, that doesn't necessarily imply that an equal water equivalent of snow would be generated at a higher elevation / lower temperature: indeed, in reality, there might not be any precipitation falling at that higher elevation / lower temperature.

So as long as others agree, I will stick with what I have now. I want to do some more testing, then will send out the branch for others to try out.

Bill S

Bill Sacks wrote on 12/5/18 10:46 AM:
Actually, I think I see a way around this one:

(i) We use different parameters for the precipitation repartitioning over glacier vs. non-glacier columns. In step (a), we are working on the grid cell mean, so which parameters should we use?
so that we can continue to use consistent, landunit-specific parameters.

Bill

Bill Sacks wrote on 12/5/18 10:09 AM:
Hi all,

While writing my commit message documenting the P1 modification (in cold temperatures, make rain run off rather than converting it to snow), I realized a possible unintended consequence: The rain-snow repartitioning has two purposes:

(1) Downscaling to elevation classes: changing the balance between rain and snow for different elevation classes

(2) Correcting problems in CAM

We've been focused on (2), but I don't remember hearing any discussion of the implications for (1), and I'm wondering if that is an unintended consequence of this change.

For example, imagine a grid cell where CAM is generating rain at a grid cell mean near-surface temperature of 3°C; CAM's topographic height here is 1000 m. On the CLM side, we have two elevation classes, one at 0 m (50% of grid cell) and one at 2000 m (50% of grid cell). In the 2000 m elevation class, the downscaled temperature will be –3°C. This will result in all rain being converted to snow.

However, with the new formulation, in the higher elevation class, rather than receiving precipitation in the form of snow, it would instead receive no precipitation! To me, this feels really bad, and really hard to justify in a paper.

A possible way around this would be to separate pieces (1) and (2), with an algorithm like this:

(a) Repartition grid cell mean precipitation using grid cell mean atmospheric temperature. If this would result in rain-to-snow conversion, instead make that amount run off.

(b) Downscale precipitation to columns, repartitioning using downscaled column atmospheric temperatures. At this stage, any rain-to-snow conversion would remain snow (as it has in the past).

I think it wouldn't be too hard to implement that algorithm: it would probably take an extra day or two for me to implement and test. However, there are a couple of issues that immediately come to mind:

(i) We use different parameters for the precipitation repartitioning over glacier vs. non-glacier columns. In step (a), we are working on the grid cell mean, so which parameters should we use?

(ii) This has the potential to give fairly different results from what was tested for P1. In some instances (like the above example), we'll end up with more snow; in other instances we could end up with less snow than with the original P1 implementation.

Please let me know your thoughts on this.

Bill S



You may unsubscribe from Associated's electronic communications<mailto:unsubscribe at ae.ca?subject=Unsubscribe+from+AE+-+CASL> at any time.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cgd.ucar.edu/pipermail/liwg-core/attachments/20181206/e4fcbd0c/attachment-0001.html>


More information about the Liwg-core mailing list