interpolation strategy to ensure geospatial accuracy #25
Description
Some of the models that we deal with are defined on regular longitude latitude grids within netcdf files. specfem_tomo_helper uses scipy's RegularGridInterpolator which is designed to work on rectilinear grids in Euclidean space. The geospatial accuracy of using this interpolator on input data defined on a regular longitude latitude grid is expected to be affected the most for large regions and regions near the poles. I'm presently working with a model for Alaska which is close to the poles, and was concerned if this might be altering my input model to specfem in an unintended way.
A better strategy might be to project the regular longitude latitude grid on to a UTM grid which will turn out to be irregular, and then use scipy's griddata which can deal with irregular grids defined in Euclidean space to interpolate on input data.
The computational time difference between interpolating on a regular grid and an irregular grid for our cases is practically negligible.