Open mcgarth opened 5 years ago
I changed the interpolation to nearest neighbour because I think it makes sense, particularly for phase data, where the geocoded output will be sub-sampled before further analysis (e.g. PyRate) and phase closure loops checked. A smoothing introduced during interpolation with another method could add noise to the phase closure loop check.
The plot thickens...
In the reference manual for geocode_back GAMMA recommend the following for MLI images:
The interpolation should be performed on the square root of the data. A mid-order (3 to 5) B-spline interpolation is recommended.
In 1d9e03d I changed the code to use option 5 (B-spline on the log of data) with order 5 (default) for MLI products (e.g. sigma0/gamma0)
Interpolation during back-geocoding was changed to nearest neighbour at e68ded4f77a4b4f641860f7383ea971b65e354ff
In the latest commit e8ca42fb828055c61b0cfed29c9f13135c616d3e this has been reversed back to bicubic spline.
We need a good reason to choose one method over another. Please add here your observations and we can come up with a decision.