Enabling stochastic parameter perturbations in microphysics

[wjcrawford]

Does anyone have experience enabling stochastic parameter perturbations (SPP) within the microphysics?

I've had success enabling sections of the SPP code, but I don't see where/how the SPP weights (spp_wts_mp) are to be calculated within the Thompson microphysics.

The Stochastically Perturbed Parameterizations section of this page (https://ufs-srweather-app.readthedocs.io/en/latest/UsersGuide/CustomizingTheWorkflow/ConfigWorkflow.html) makes it look like the flags need to be defined in a config.yaml file, but I'm unsure where the file should live, and if I need to build it myself.

Thanks for any help you can provide!

[grantfirl]

@wjcrawford Could you provide some info about which host model/application you're wanting to use SPP with? Alex Reinecke suggested that it was with the SCM, and I don't think anyone has done much with the SCM and stochastic physics in general. If you're using some flavor of the UFS besides the SCM, there may be better places to put this question, like the stochastics_physics repo (https://github.com/NOAA-PSL/stochastic_physics) on GitHub? Let me know if you're trying to use the SCM with this and I can look into understanding how to generate SPP weights for it.

[grantfirl]

You're definitely "breaking ground" if you're using the SCM with stochastic physics. I'm curious as to how you'd like to handle the spatial perturbation component for a single column. Would you like the SCM to be able to have a "ghost" grid around it with a spatial pattern and the column of interest gets updated values as the pattern changes in time? Or, is this a situation where you've run the UFS and want to run the SCM retroactively using UFS forcing with the SPP weights that a particular column was given? Or something else?

Up to now, as far as I know, the SCM has avoided adding any of the stochastic physics code because of not knowing how to handle the horizontal spatial aspect, or what would even be useful within an SCM framework. Let me know what you're thinking of doing and we may be able to provide pointers, but I doubt that we'd be able to help much with actually writing an implementation or adding this feature to the SCM on our side.

[JeffBeck-NOAA]

Grant laid out the issue with using stochastic physics in the SCM really well. It depends on what the goal is. If there's an interest in mimicking the use of stochastic physics in the SCM to make it behave the way it does in a 3D simulation, then we need the ghost grid that Grant defined, which is controlled by spatial decorrelation lengths and times. Even if there's no interest in following the same stochastic physics pattern evolution that, say, the UFS uses, there would still need to be a method through which the perturbations vary in time. It could be really basic, such as a sine/cosine function with some added noise, or it could be more complicated.

[wjcrawford]

Thank you both for your replies. I spent some time thinking about it, and really I'm not that concerned about the spatial variation at this time. Mostly, I want to see what happens to various cloud/precipitation variables within the SCM when the microphysical parameters are perturbed, both one at a time and with combinations of perturbed parameters. In which case, I'm thinking we would not need to deal with a ghost grid. As you note Jeff, we would need some way to very the perturbations in time, and to start with a simple sine or cosine function would suffice, which I hope helps to simplify the problem.

[wjcrawford]

Hi @grantfirl and @JeffBeck-NOAA, just wondering if y'all have any additional thoughts on adding SPP perturbations in the SCM with a simple time variation? Eventually we'll want to move toward spatially varying perturbations, but to start it would be nice to get an idea of how impactful the perturbations are in a more controlled environment

[dustinswales]

@wjcrawford As Grant mentioned, there is no infrastructure to enable stochastic physics within the SCM; However, it would be straightforward to build, assuming you had some kind of a "ghost grid" as Grant mentioned (great name btw).

For temporal perturbations, I believe you would only need to:

• Compute the time-varying pattern at the same spot as the spatial-varying in the atmospheric driver (e.g. UFS/SCM atmospheric drivers linked below). Maybe this could be built into stochastic_physics_wrapper as an option? Paving the way for future ghost-grids for spatial stochy in the SCM?
• Then pass these temporally varying weights into the physics, where they will be applied within some scheme(s).

In the UFS, the stochastic physics patterns are generated just after calling the CCPP timestep_init phase. Assuming you had the grid-info, the call to the pattern generations could be added to the SCM in the same spot.