Hydro model channel improvements #156
Conversation
Desirable for AIS or for the situation where we do not consider thermodynamics.
When externally provided, these melt inputs will not change during the SGH subcycling, so they don't need to be evaluated within the subcycling loop. This also moves one halo update out of the subcycle loop.
Useful for profiling where performance gains might occur
This commit fixes a bug that was introduced when downhill terrestrial outflow was modified to ignore to the downhil bedtopo slope. This commit ensures that there is no outflow if the bedtopo external to the grounded ice is a higher elevation than the bedtopo under the ice. This was previously occurring.
This seemed to be a failed experiment as it prevents channels from ever getting very extensive. I'm keeping the previous commits in the history for posterity.
In AIS simulations, having iceThicknessHydro enabled was causing massive lake formation and lack of channelization. On further investigation, I found there are thickened artifacts of increased thickness adjacent to grounding lines. This was in effect creating berms in the hydropotential, making it harder for water to reach the ocean. The iceThicknessHydro calculations should be improved to eliminate this issue, or, if that's not possible, removed from the code. But for now I am disabling it by default. Note that I created an MPAS-Tools script to preprocess ice thickness to eliminate depressions in the surface elevation, and this greatly improves model stability around subglacial lakes. That preprocessing step should be used instead of iceThicknessHydro. It is unclear if that will be sufficient in the case of evolving ice thickness. If not, the surface-depression-filling algorithm it employs should be ported to MALI and replace the existing iceThicknessHydro algorithm.
matthewhoffman
added
bug
|
Test run with input files is here on perlmutter: Mass balance & time series from plotting script in MPAS-Dev/MPAS-Tools#520 : Comparison of water thickness within 1% of bump height (light shading) compared to radar specularity for Thwaites: Comparison for more of WAIS (obs from https://royalsocietypublishing.org/doi/full/10.1098/rsta.2014.0297), lakes in peach, water thickness within 1% of bump height in light shading: |
alexolinhager
left a comment
alexolinhager
left a comment
There was a problem hiding this comment.
Thanks for addressing these issues with the hydro model Matt! I made a few minor comments/suggestions, but otherwise should be ready to merge
| channelMelt(iEdge) = 0.0_RKIND | ||
| channelPressureFreeze(iEdge) = 0.0_RKIND | ||
| endif | ||
| enddo |
There was a problem hiding this comment.
We had previously tried something similar to this (E3SM-Project@c01e547), but had done so by disabling all channel evolution instead of just disabling channelMelt and channelPressureFreeze over lakes. The other attempt never got merged because we still couldn't get stable channels, but this seems like a better approach, especially when paired with the other commits in this PR
There was a problem hiding this comment.
Reading two more commits down the line, I think the issue with this approach is familiar and seems to be
why we discontinued it previously
|
|
||
| <nml_option name="config_SGH_use_iceThicknessHydro" type="logical" default_value=".true." units="unitless" | ||
| description="Option to use an altered ice thickness field called iceThicknessHydro that replaces local maxima/minima in upperSurface with a mean of the cells neighbors. This option has no significant effect on the behavior of the model but makes it more stable." | ||
| <nml_option name="config_SGH_use_iceThicknessHydro" type="logical" default_value=".false." units="unitless" |
There was a problem hiding this comment.
I think we will eventually need to port your MPAS-Tools ice surface alteration code into MALI if we want to run with evolving dynamics. The primary reason for iceThicknessHydro was the development if single-cell surface depressions that formed when running coupled simulations. This was mostly an issue when running with the higher order velocity solver, which created speckled ice thicknesses along the domain edges as the run progressed.
It seems like the MPAS-Tools code should fix that issue, so I think we should try replacing the iceThicknessHydro code with what you wrote there. Would it be better to make that it's own PR, or tack it onto this one?
| channelArea = channelChangeRate * deltatSGH + channelArea | ||
| ! If sheet dissipation contributes to channel, there should be no need for a minimum channel size | ||
| channelArea = max(1.0e-8_RKIND, channelArea) ! make some tiny value when it goes negative | ||
| channelArea = min(config_SGH_chnl_max_area, channelArea) ! limit channel area to config-specified max value |
There was a problem hiding this comment.
Would it be better to just limit channelMelt and channelPressureFreeze when channelArea exceeds config_SGH_chnl_max_area? That way the channel still wouldn't grow past the max value but we also wouldn't violate mass conservation if there was no actual unaccounted melting. Probably doesn't affect the outcome but could be cleaner to not have a cell where channelMelt far exceeds the channelArea
This pull request introduces a number of developments to get the subglacial hydrology model running reliably with channels active.
Most significant changes are:
Less important changes are: