scale_region strange behavior

[phillipyeh]

I started with the 92-25 km circular mesh, centered the higher resolution portion over the Northeast US (40N, 73W), cut out a circular region with radius 17000 km, and then used the scale_region.py tool to scale it down by a factor of 8.

I then ran a 24-hour simulation starting 18UTC on 16 Feb 2024. I was able to successfully generate output files, with no error messages in the log files. However, when I look at the output, it does not match the observed output at all. In particular, over the Eastern US, the precipitation is in the wrong location, and the pressure/height fields do not match observations.

I checked the land-ocean mask in the history* files, and the mask is in the correct place. So the issue does not appear to be with the grid.nc or static.nc files.

The grid files and output are located in this cloud directory:

Box

stonybrookmedicine.box.com

Let me know if you need anything else. Thanks!

 

[mgduda]

Could you post a plot or two here to give an idea of how badly the simulation results are mismatched with observations? Certainly, the simulation results will not be a perfect replication of reality, so some mismatch in the location of precipitation, and some disagreement in the pressure and height fields is probably not unexpected.

The scale_region.py tool uses a stereographic projection to map the regional mesh from the sphere to a plane, where it is scaled by the prescribed factor. With a radius of 17000 km, the unscaled regional domain would cover more than a hemisphere of the globe, and perhaps the large map scale factors in the projection of that regional domain onto a plane are leading to issues: when scaling the mesh on the plane, cells in the mesh are translated between regions of different map scale factor, leading to differences in the scaled cell areas from what one would expect based on an application of the nominal scaling factor.

 

[phillipyeh]

Attached are SLP + radar composites at 18UTC (19UTC in MPAS) on 16 Feb, 00UTC on 17 Feb, and 06UTC on 17 Feb. While the fields at 19UTC still look somewhat close to the WPC analysis, by 06UTC the model has diverged significantly from the WPC.

If the larger-than-hemisphere unscaled domain leads to large map scale factor issues, is there another way to generate a mesh of this size that scales down from ~12km to ~3km?