Add IRI #3
Add IRI #3
Conversation
|
I go through the steps for processing the ASAP phenology in I'm also using the outputs (in |
There was a problem hiding this comment.
Okay, not so many comments here as obviously this overlaps a lot with the Sahel codebase, we aren't going to use the livelihoods zones or other admin stuff for now, and the full processed code will be ported into the global monitoring repository.
I put a few comments in the code, but I think here are the overall comments:
- Just general comment sin the file, to explain what's being looked at and also just anything interesting or useful found there. Can help in the future when returning to these and other people or even yourself or trying to parse exactly the reasoning for doing certain things.
- In this vein, I think just a simple README.md explaining what was explored here would be good, and mentioning the files that are there and maybe a quick explanation of what was done or what to look at.
- For this, I think we should just have top level folders for different datasets being explored. Otherwise, this repo will be way too cluttered. For this, I think each folder should be self-contained and require nothing from outside. Basically a way to put everything in one location but without requiring these modules work together. So it would be
asap/src,asap/explorationand the read me would sit under the folder asasap/README.md. Then in the overallREADME.mdcan just have a line or two explaining what is in theasapfolder with a link to the folder itself (so theasap/README.mdwould be displayed on the GitHub platform. - In the future, might be better to split out the utilities. Not really mattering here, but for instance in the Sahel repo would make it cleaner to read through and connect which work together and which provide disparate functionality.
- No requirements file. Not a big deal, but probably best to provide at least. You could use
pipreqsto generate the file for theasapfolder, and store under there.
| """ | ||
| filepath = ( | ||
| DATA_DIR | ||
| / "public/processed/glb/acaps/seasonal-events-calendar_processed.csv" |
There was a problem hiding this comment.
Likely never an issue, but if you separate all folder and file names as individual strings, they will be fully system independent pathing using pathlib.Path, but currently this still assumes the / path separator.
src/utils.py
Outdated
|
|
||
|
|
||
| def process_drought_codabs(): | ||
| """ |
There was a problem hiding this comment.
For exploration this is fine, but for eventual bringing into global monitoring we won't want to request to countries with AnticiPy config files. If we need admin1 at the country level, much easier to use something like https://fieldmaps.io/data which is already aggregated globally and was based off the original CODAB files so in most cases should match the names.
| # join with CODAB | ||
| # note - some asap1_ids don't match up, | ||
| # hence will be missing from plot | ||
| cod_crop = cod_asap.merge( |
There was a problem hiding this comment.
cod_asap is not defined in this notebook.
|
|
||
| ```python | ||
| lon, lat = 0, 0 | ||
| da.where(da < 251).sel(x=slice(lon, lon + 40), y=slice(lat + 40, lat)).plot() |
There was a problem hiding this comment.
I know these are just autoplots, but would be good to hear to have a single color band, which would make it easier to see where the intensity lies. And is it the case that this is the # of dekads in 3 months that are in season? if so, there seems to be a large swath with just 1?
| # 253 = hard to tell, barely used | ||
| # 252 = hard to tell | ||
| # 251 = no season (desert but also rainforest?) | ||
| lon, lat = -68, -14 |
There was a problem hiding this comment.
Maybe just a comment on where we are looking at.
|
|
||
| ```python | ||
| # plot cumulative distribution of probability (reversed) | ||
| df.hist("prob", cumulative=-1, bins=100, density=1) |
| geobb = GeoBoundingBox.from_shape(adm0) | ||
|
|
||
| ds_adm0 = ds_f.rio.clip(adm0["geometry"], all_touched=True) | ||
| # resample to 0.01 degrees |
There was a problem hiding this comment.
Question, what's the point of the resampling here instead of just using the base IRI output? Might be good to make clear in the analysis files. Is this something we want to do in the full implementation?
| da_d = da_d.rio.set_spatial_dims("x", "y", inplace=True) | ||
| # da_d = da_d.astype("uint8") | ||
| for dekad in dekads: | ||
| da_d.loc[{"dekad": dekad}] = ( |
There was a problem hiding this comment.
Not sure if more efficient, but might be able to use the modulo operator I referenced in the other PR to reduce this to just the 2 sections connected by | for season 1 and 2.
|
|
||
| # save | ||
| # File too big to save as 36-band raster, so must save as multiple files | ||
| # Note that saving as an actual Boolean in a NetCDF is even bigger somehow. |
Key processing steps:
explorationfolder, but can be moved tosrcif we decide we're happy with this method.src)country_configin AnticiPy, and are included in the ACAPS dataset (insrc)Also includes
approx_mask_raster()to up-sample Xarray datasets.Also includes notebooks in
explorationfor exploring:All very much preliminary stuff, but I imagine we'll be using the seasonal calendars soon so thought you'd want to get a look at the processing.
Also, branch name is definitely too broad in scope! Sorry