Skip to content

Demonstrate CDO wrapper transform in pyearthtools pipeline #247

@stevehadd

Description

@stevehadd

CDO tools are widely used by climate science and well understood. Being able to use CDO through a pyearthtools pipeline which can than easily integrate into a ML framework (e.g. pytorch), would be a really good way for climate scientists to make use of existing skills as part of an ML project.

We shoukd demonstrate

  • a pyearthtools accessor to some climate data e.g. CMIP6 models
  • extract variables with CDO cdo selname, pressure levels etc. using CDO
  • extract a region using CDO cdo sellatlonbox
  • possibly calculate daily means cdo daymean
  • rename to CF names cdo chname,old_name,new_name
  • regrid to coarser resolution using CDO cdo remapcon,outgrid.txt -setgrid,ingrid.txt infile.nc outfile.nc
  • return pairs of coarse predictors and high-res target(s)
  • train basic CNN downscaler (use current tutorial)
  • inference
  • evaluate with scores package
    • Look at monthly and yearly means of variables to check whether climatology correctly represented. cdo monmean , cdo yearmean
    • Calculate climate indices with spaital averages cdo fldmean or cdo fldmax

Some example material

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions