Use closest year available for each map

[hannah-rae]

For maps that have more than one year available, we use the closest year to the test set year.

TODO: update results for all map notebooks

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ivanzvonkov]

I don't really love that every time an intercomparison is made this file needs to be updated. I think it would be cleaner to minimize (or eliminate) changes to this file, and set the configuration (dataset name, admin code, year) in each respective notebook. In my view it'll be a bit more modular and pleasant to use that way.

Of course that script that generates all the intercomparisons would need to be updated or deleted, but since this is the primary way we are doing intercomparisons, I would be alright with that.

What do you think? @hannah-rae

[ivanzvonkov]

Should this parameter be a list to allow sampling from a range of years?

[ivanzvonkov]

Might be a more explicit way of specifying which maps should or should not be sampled

[ivanzvonkov]

This collection only includes images for 2019

[hannah-rae]

Oh yeah, this one is weird. I'll make it so that it changes the collection.

[ivanzvonkov]

Ah this filter date thing is a bit messy now with the years. What do you think about making it a regex style in all instances.
.filterDate("YYYY-01-01", "YYYY-12-31") and that YYYY is replaces with the year selected?
Also filterDate is the same as .filter(ee.Filter.date so it can be replaced everywhere

[hannah-rae]

I think that may work... the issue I had when thinking about this is that the Covermap objects get created before they are sampled. But as long as every dataset has a year specified then it should be fine.

[ivanzvonkov]

Evaluating the ee_asset from a string can just be moved into extract_test to fix this:
ee_asset = eval(ee_asset_str.replace("\n", "").replace(" ", ""))

[adebowaledaniel]

Updates:
compare_covermaps:

• the get_test_years gets rid of the TEST_CODE, per Ivan’s review
• Added the year to the names of the Togo, Rwanda, and Uganda test files to match the name convention. (I will update the actual files on DVC later.

These changes were made to ensure the year parameter is maintained in the extract_test and to prevent any errors in its use case in TestCovermaps.

Demo-notebook (to be deleted before merge):

• Without a range filter of maps, I observe that the extract_year only selects the nearest year from maps with multiple years_covered and updates the ee_asset string. For other single-year maps, we could select maps with the exact test year or those within a range of ±1 year of the test year. (The latter is implemented in the notebook). The analysis will focus on maps produced closest to the test set year.
  This might be a use case specific to area estimates; however, let me know what you think about this for intercomparison @hannah-rae. I will include it in the area estimates notebook in a separate PR.