CCI-Tools / cate

ESA CCI Toolbox (Cate)
MIT License
51 stars 15 forks source link

Coregister fails for non-overlapping domains #495

Open thomas-eldridge opened 6 years ago

thomas-eldridge commented 6 years ago

Expected behavior

Downloaded/ opened two datasets covering identical temporal range but different geospatial regions (i.e. reduced download size while exploring teleconnection). Expected to be able to coregister the datasets (to same lat/lon granularity) despite lack of geospatial overlap.

Actual behavior

Error raised by Cate, presumably because latitude/ longitude domains do not overlap

Steps to reproduce the problem

  1. Download/open SST ds for 5S - 5N, 170W - 120W (NINO3.4 region)
  2. Download/open soil moisture ds for Northern AUS/ Malaysian peninsula (because known teleconnection w. ENSO)
  3. Attempt coregister

Specifications

Cate v1.0 dev 3

forman commented 6 years ago

Does the same apply to Cate 1.0? Note that version 1.0 dev 3 is outdated.

JanisGailis commented 6 years ago

Yes, the coregistration operation has been defined Coregistration op specification in the context of Use Case 9 to enable pixel-by-pixel analysis of two datasets observing the same spatial scene. Hence the grids include spatial extents as well, and even do cropping if necessary.

@thomas-eldridge Is pixel by pixel analysis important to explore teleconnection? @forman It should be possible to change how the operation works by relying on pixel size alone to determine width and height of the slave dataset, such that regions that don't overlap can be coregistered. It will mean that how coregistration operation works will change, so on top of that we might need to implement subset_as to preserve the current functionality. What do you think?

forman commented 6 years ago

I'd prefer, if coregistration worked like @thomas-eldridge is expecting it to work. Btw, this is also how the SNAP collocate operator is working.

What about a new keyword argument allow_disjoint_inputs: bool = False.

JanisGailis commented 6 years ago

@forman In your opinion, should coregistration crop the slave to master as it does now?