Preprocessing for pkg/profiles and pkg/obsfit

Authors

Goldberg, M., Griffin, C., & Reich, S., Processors, LLC.

Background

MITgcm allows users to compare data with model equivalent information, often for the purpose of optimization of a misfit. There are three primary packages used to accomplish this task:

1. pkg/ecco: Used for gridded data, i.e. datasets that have been mapped to a MITgcm model grid
2. pkg/profiles: Used for ungridded profile data. When dealing with sparse ctd data, embedding profiles into gridded fields a la pkg.ecco results in needlessly sparse fields. To lessen the memory burden, pkg/profiles was developed to interpolate profiles or vertical columns of data into the model grid for model-data comparison
3. pkg/obsfit: In the case in which the user wants to compare pointwise ungridded observational data defined by their latitude, longitude, and depth, it may be wasteful to use profiles. pkg/obsfit was created for these cases.

Files and folders in your project repository

Collaborators

Name	Personal goals	Can help with	Role
Caeli G.	Increase Python confidence	Reading/translating MATLAB scripts	Project Lead
Matthew G.	Write modular open source code	Providing template routines to build off of	Project Lead
Shoshi R.			Project Lead

The problem

Presently, there are no known (to the authors) publicly available preprocessing routines for pkg/profiles and pkg/obsfit written in python. Gael F. has a public set of routines in MATLAB, which serve as inspiration for this project. Given that the ECCO community has embraced python in recent years, in particular harnassing xarray, we see the addition of a python data preprocessing module to be an attractive contribution to the community.

Project goals and tasks

Project goals

• Produce a new python package to preprocess profiles and obsfit data
• Create jupyter notebooks demonstrating different uses of the package, examining different routines within the package

Tasks

• Write module obsprep to prepare obsfit and profiles objects, assembling interp information (including grid='llc')
• Investigate nearest neighbor methods -- KDtree is not innately aware of spherical geometry -- use pyresample
• Finish documentation, tests & installation scripts
• Set pkg='profiles' default prof_interp_weights to be given by bilinear interpolation coefficients following Gael's routine. In summary: final nearest interp points using new sphere-aware search via pyresample. Map interp points from spherical->cartesian, orthogonalize grid cells, then compute coefficients.
• Default for interp_weights for pkg='obsfit' should be trilinear, following a similar routine.

Notes:

• Interpolation: profiles by default does bilinear interpolation, computing the model equivalent to an ungridded observation as a weighted average of its four neighbors, from wikipedia).

Future additions

• Put nearest neighbors search mapping ungridded -> gridded inside MITgcm
• Examine treacherous interpolation at the poles

Project Results