Clip using geodataframe and rioxarry.clip not clipping xarray dataset

Question

I am trying to clip an xarray dataset using rioxarray.clip with geometry from a geodataframe. I am running the code on a cloud platform built on kubernetes clusters in a distributed manner using Dask. From what I can tell, my code is correct and should be performing the clip. However, when I graph the clip result the clipped dataset has the same extents as the parent.

First, the data is brought in using the intake library and the to_dask function.

url = 'https://raw.githubusercontent.com/USGS-python/hytest-catalogs/main/hytest_intake_catalog.yml'
cat = intake.open_catalog(url)

dataset = 'conus404-40year-cloud'

ds = cat[dataset].to_dask()

This is what the dataset looks like. The dataset is then trimmed to variables and time of interest. The clipping geography is also created.

# restrict variables to those of interest
c404_variables = ["PREC_ACC_NC", "TK", "ACSWDNB", "I_ACSWDNB", "ACSWDNB", "I_ACSWDNB", "ACLWDNB", "I_ACLWDNB", "ACLWUPB", "I_ACLWUPB"]
c404 = ds[c404_variables]

# use slice to filter by time 
start_time = "1990-01-01 00:00"
end_time = "1990-12-31 00:00"
c404 = c404.sel(time=slice(start_time, end_time))

#WGS84 = EPSG:4326
c404.rio.write_crs("epsg:4326", inplace=True)

# bring in boundaries of the Deleware River Basing (DRB) and create single polygon
drb = pygeohydro.WBD("huc6", outfields=["huc6", "name"]).byids("huc6", ["020401", "020402"])
# create a column where all entries have the same value
drb["name"] = "DRB"

#dissolve by that column to create a single geometry
drb = drb.dissolve(by="name")

At this point, I've matched the CRS between the xarray dataset and the geodataframe. I then perform the clip and visualize the result.

# clip data to precise geometry of Delaware River Basin
c404_drb = c404.rio.clip(drb.geometry, drop=True, invert=False, from_disk=True)

# make data smaller 
c404_drb_check = c404_drb["TK"].sel(time="1990-01-01 07:00").load()

c404_drb_check.hvplot.quadmesh(x='lon', y='lat', rasterize=True, 
                             geo=True, tiles='OSM', alpha=0.7, cmap='turbo')

And the resulting image has an extent of CONUS rather than the Delaware River Basin.

After reviewing the rioxarray usage examples and API references, I wonder if there are issues with the coordinates of the dataset or that it was created through intake.to_dask, though it is an xarray.dataset so rioxarray should still work. Any insight someone would have would be most helpful. TYIA

Answer 1

rioxarray clip has not been designed/tested to work with dask yet. It may or may not work. If you are interested in collaborating to add support, opening a ticket on GitHub would be a good next step.