Continued work to wrap up north-up corrections

[nwagenbrenner]

Remaining TODOS from #544

• need to double check wrfSurfaceInitialization.cpp, I saw an error claiming the coordinateTransformationAngle was not able to be calculated. I suspect some of the manually created gdal dataset stuff, like projection reference or geo transform, are not being set correctly for it? Related: I also noticed that many of the, return false with no error message, code in the gdal_util.cpp functions, was not helpful when checking this type of calculation failures. Better error messages probably need to be added where able.

• all the standard initialization methods were tested for utm12N locations warped to various projections, but still need to test across various places on the globe. So need to repeat at least domain average runs at 270 degrees, across various locations of the globe warped to various projections.

• figure out what is going on with the Robinson projection, it breaks the new conventions that we've come up with. Test the flat_test.tif case with more projections

• hrrr_to_kmz has an issue where the coordinateTransformationAngle is calculated using the whole CONUS dataset, rather than the clip bbox, resulting in a coordinateTransformationAngle of near zero, when it should be a lot bigger value for the region of interest.

• flow_separation_grid has yet to be looked over and checked for angleFromNorth calculations

• None of the 3D initializations have yet been looked over or checked for the angleFromNorth calculations

• there is still a question with the conventions. coordinateTransformationAngle_geog_to_dem is correct for utm18N (~-28 for big_butte_small.tif), but coordinateTransformationAngle_dem_to_geo is incorrect for utm18N (~32 instead of ~28 for big_butte_small.tif). This implies that always calculating the coordinateTransformationAngle FROM something TO dem may break down as a calculation method. I have a suspicion that the real method should always be FROM geo TO something, which is the angleFromNorth calculation. So the coordinateTransformAngle would then be calculated by adding together the angleFromNorth of the two datasets, though not sure what the various signs would need to be for such a calculation to be correct. But this hypothesis would need to be tested somehow, I have yet to see a scenario where the normal convention breaks down other than the Robinson projection case. Regardless, we need to be on the look out for situations when the current convention breaks down in this way.

• should probably add a comment here with the current conventions described very robustly, maybe even with diagrams.