Closed dwfncar closed 6 years ago
Mallory and I talked yesterday about correcting the NCEP subregion masks in MET to fix the discrepancies she found.
The NCEP masks are defined on grid 104, but the VSDB code essentially regrids them to the evaluation grid on the fly. It occurred to me this morning that we could make MET do something very similar.
Currently, the MET tools require that gridded mask fields must be on exactly the same domain as the verification domain. You often do this by running gen_vx_mask with a polyline file. If the grids don't match you get an error message similar to:
"The masking and verification grids do not match"
At this point in the logic, rather than erroring out, I could have MET automatically regrid the mask to the verification domain using nearest neighbor interpolation. That would mimic the VSDB behavior.
Michelle and Mallory agreed this would be good functionality. Make the log message DEBUG(2) so it shows up in the output by default.
by johnhg
Finished updating the NCEP polylines.
Along the way, found a bug in MODE that it couldn't make PostScript plots when match = NONE. Fixed that.
Also, removed all references to the zero_border_size MODE option since we removed that as a requirement.
by johnhg
I committed the updated versions of the polylines but they led to many differences in the regression test! The poylines which do not touch the boundary of grid 104 should not change at all. I will remake them using a custom version of MODE where we dump all points rather than only the points where we change direction. Hopefully that will produce better results! by johnhg
After including all points in the lat/lon polylines (not just the vertices), the number of points in the polylines matched up. But the lat/lon values are slightly different... up to about 0.3 degrees, which is pretty big.
I tested calling "single_boundary_offset(true, 1, 0.5)" to see if the lat/lon value was the center of the grid box rather than the lower left corner. But that only made the differences worse.
Next, I ran METv1.1 pcp_combine and the met_6.1 and looked at the grid defs. Version 1.1 had the radius of the earth set as 6367.47 while version 6.1 uses 6371.20. I updated 6.1 to use the old radius, but we still have significant differences.
by johnhg
I ran grid_stat using the old polyline definitions, new polyline definitions, and automated regridding. I applied this to GFS data and looked at the number of matched pairs. For each region, the TOTAL count is listed for the old polylines, new polylines, and NCEP mask. There isn't much of a consistent pattern here!
APL 164 160 163
CAM 399 435 431
CAR 850 881 876
CONUS 3523 3569 3520
EAST 1670 1692 1674
ECA 2552 2550 2550
GLF 547 543 541
GMC 195 196 196
GRB 200 203 198
HWI 45 45 46
LMV 329 337 330
MDW 609 622 609
MEX 876 887 879
NAK 902 929 902
NAO 4610 4793 4805
NEC 177 181 177
NMT 373 374 372
NPL 372 383 367
NSA 304 362 361
NWC 151 154 149
PRI 99 101 102
SAO 2415 2688 2691
SEC 196 196 199
SMT 229 215 232
SPL 306 319 305
SWC 77 85 78
SWD 145 144 145
WCA 2555 2587 2552
WEST 1853 1877 1846
by johnhg
We have switched to using Shapefiles instead, so these Grid 104 definitions are deprecated. by johnhg
This task is the result of discrepancies found by Mallory Row in her application of MET to global datasets. This task is to remake the NCEP regions using MODE but making sure not to zero out any points along the boundary.
/scratch4/NCEPDEV/global/save/Mallory.Row/VRFY/vsdb/exe/grid#104
Could include a note as a NetCDF attribute describing how it could be run through regrid_data_plane using nearest neighbor interpolation to put it on another grid.
[MET-878] created by johnhg