GCM munging - address NA cells

[hcorson-dosch-usgs]

Alright - I have a solution in place for the missing data, which fixes #296

Per Lindsay and my discussion, we landed on an approach that did not change any of the data for cells with missing data, but instead adjusted the lake-cell-tile xwalk so that lakes were matched to only those query cells that returned data, on the basis of which query cell centroid was closest to each lake centroid.

To avoid having to read in the data twice (once in order to determine which cells are missing data, and again to munge the data for cells that are not missing), as I proposed in the 2nd bullet under considerations in #296, Lindsay suggested that I try returning a list of two items from munge_notaro_to_glm():

• file_out - The name of the output file of munged data for the given gcm and tile
• cell_info - A tibble with a row for each cell in the given tile, with T/F values for whether or not that cell was missing data

That seemed like a great approach so I set up the targets workflow like so:

• In place of glm_ready_gcm_data_feather have munge_notaro_to_glm() instead build the target glm_ready_gcm_data_list, which has a branch for each tile-gcm combo, and 2 elements for each branch: file_out and cell_info, as noted above.
• Then construct glm_ready_gcm_data_feather from the file_out element of each branch. _Note: this ended up requiring mapping over glm_ready_gcm_data_list, as each branch of that upstream target had the two distinct elements, and it was not possible to pull just one element from each branch in one go with glm_ready_gcm_data_list$file_out_
• Similarly, construct a combined tibble, glm_ready_gcm_data_cell_info from each of the cell_info elements of the branches of glm_ready_gcm_data_list. _Note: again this required mapping over the branches of glm_ready_gcm_data_list_.
• Then, once I had that combined tibble, I was able to build glm_ready_gcm_data_cell_status, a pivoted version of glm_ready_gcm_data_cell_info with a single row per cell, and columns indicating whether or not data was missing for each gcm as well as a final column with the total count of gcms for which data was missing.
• Then I used that glm_ready_gcm_data_cell_status to build the new revised lake-cell-tile xwalk lake_cell_tile_xwalk_data_df (the original is renamed to lake_cell_tile_xwalk_spatial_df) matching lakes to only those query cells that returned data, based on which query cell centroid is closest to each lake centroid.
  • The lake-cell matching will only differ here from that in lake_cell_tile_xwalk_spatial_df for those lakes that fell within cells that were missing data. I checked and, as expected, the two crosswalks differ for 123 lakes, which was the number of lakes that fell within cells with missing data
• This lake_cell_tile_xwalk_data_df now gets saved as lake_cell_tile_xwalk_csv to be used in lake-temperature-process-models to determine what meteo data to pull for each lake.
• I also added a target to map the new lake-cell-tile xwalk

Mapping with the spatial xwalk - represents the number of lakes in each cell used in the GDP query:

Mapping with the revised xwalk - represents the number of lakes in each cell that will use data from that cell as meteo data for GLM modeling

And the map of missing cells with nearest cells to lakes within missing cells noted in green:

While I was in the code I also cleaned up some of the function documentation and made the use of cell_no consistent throughout (some of the munging code used cell instead).

[hcorson-dosch-usgs]

I did build all of the downstream targets again to ensure everything built, but I am just now building out_skipnc_feather to ensure it works with the change I made from cell to cell_no. It's a slow target to build so I didn't want to hold this PR. I'll comment once it's built re: whether or not it built successfully.

[hcorson-dosch-usgs]

Hmm, well. I left it running overnight and it stalled, but it did build 6 of the branches first, so I think it would work. I did try restarting it, but it's hanging on that same branch.... Trying again now with error = 'continue' to see if that branch actually errors/ if other branches will build.

[hcorson-dosch-usgs]

Also - I was just scrolling back through our chat, Lindsay, and noticed you'd suggested having a single xwalk with a cell_no_spatial column and cell_no_data column. In my head I'd interpreted that as two xwalks, so that's what I currently have in the pipeline, but I could combine them into a single crosswalk if you think that would be better.

[hcorson-dosch-usgs]

FYI all branches of the out_skip_nc target did build