Closed dzaugis closed 3 years ago
Here is a screen shot of inverse distance vs bilinear interp of a rectangular grid. The original grid is top left, inverse dis top right, bilinear to 1x1 bottom left, bilinear to 1/4 bottom right
And here is a shot of interpolating a tripolar grid
The original grid is top left, inverse dis is top right, bilinear is bottom left, and the example interpolation given by Jamie and mike is bottom right. The top right and bottom right are identical
The bilinear looks better along the coast but misses so areas in the arctic ocean.
You are flying through this stuff! Yet another one of these great questions, not sure as if I have an answer. My gut reaction is to go with what Jamie and Mike are doing. Though, they generally have a "global" interest where as we are really going to be focused on the northwest Atlantic.
I guess one thing to think about (and maybe we can chat about it Monday at some point), is how different the values across methods would be if we were to crop each of these examples to the NELME shapefile (so ignoring all values over land)? I zoomed in a bit and the values seem somewhat similar even though the inverse distance weighting results in more values encroaching over land. @kemills might have some thoughts here too...
Thanks for the examples to look at, Matt. My thought also would be to go with what Mike and Jamie are doing, so if we need to fill any gaps later or rely on them for anything, our methods would be the same. On the land issue, I was thinking similar to Andrew I think in that I was assuming at some point land gets cropped out, so the over-land interpolation would be remedied with that step...but maybe that wasn't a step you were viewing as necessary? Let's discuss tomorrow.
Interpolation decisions- Interpolate: inverse distance weighting method Function: CDO remapdis Reasoning: Consistent with the same methods as Jamie and Mike from ESRL
Below is an overview of different interpolation methods and their pros and cons. No single method works perfect for all of the different grid types.
Inverse distance weighted: Pro: Used by Jamie and Mike to interpolate to a standard 1x1 degree grid. Pro: Works well for curvilinear grids Con: Some stretching occurs when interpolating different rectangular grid sizes to 1x1 deg. A fair amount of encroachment over the land occurs.
Bilinear interpolation: Pro: works best for interpolating rectangular grids to the destination grid. Con: Struggles interpolating values of tri-polar grids at the poles, often leaves NAs over the arctic ocean
Additionally, using the OISST 1/4 grid as the input file dramatically improves the coastline resolution for both inverse distance and bilinear interpolation
@aallyn do you think it is fine to use inverse distance for the curvilinear girds and bilinear interpolation for the rectangular grids?