incorrect coordinate when export ERA5L

[kongdd]

I faced with a strange issue. It seems that lon and lat is confused (lon becomes lat, lat becomes lon).

import ee
import xee
import xarray as xr
import xarray

# ee.Initialize()
ee.Initialize(opt_url='https://earthengine-highvolume.googleapis.com')


def ee_col_download_year(region, col_id='ECMWF/ERA5_LAND/HOURLY',
                         year=2022, fout="a.nc"):
    # fout = prefix + col_id.replace("/", "_") + "_" + str(year) + ".nc"
    
    ic = (ee.ImageCollection(col_id)
          .filter(ee.Filter.calendarRange(year, year, 'year'))
          .select(['surface_net_solar_radiation'])
          .limit(10)
          )
    print(fout)
    print(ic.size().getInfo())

    ds = xarray.open_dataset(
        ic,
        engine='ee',
        projection=ic.first().select(0).projection(),
        geometry=region
    )
    ds.to_netcdf(fout)
    ds

region = ee.Geometry.Rectangle(108.5, 28.75, 116.25, 33.5)
ee_col_download_year(region, year=2024, fout="test.nc")

Inspect in R:

library(terra)
ra = rast("./test.nc")
plot(ra[[1]])

netCDF4 is installed (v1.7.1).

[kongdd]

According to the comments of Alexander-Barth, (the developer of NCDatasets.jl). The netcdf that xee exported has the following format errors:

1. coordinate variable must not have the _FillValue or missing_value attributes:
   https://github.com/cf-convention/cf-conventions/blob/main/conformance.adoc#5-coordinate-systems-and-domain

        double lon(lon) ;
                lon:_FillValue = NaN ;
        double lat(lat) ;
                lat:_FillValue = NaN ;

2. the bounds attribute is expected to be a string (a containing a variable name with related metadata):
   https://cfconventions.org/Data/cf-conventions/cf-conventions-1.11/cf-conventions.html#cell-boundaries

netcdf xee-test {
dimensions:
        time = 10 ;
        lon = 78 ;
        lat = 48 ;
variables:
        float surface_net_solar_radiation(time, lon, lat) ;
                surface_net_solar_radiation:_FillValue = NaNf ;
                surface_net_solar_radiation:id = "surface_net_solar_radiation" ;
                surface_net_solar_radiation:data_type = "{\'type\': \'PixelType\', \'precision\': \'double\'}" ;
                surface_net_solar_radiation:dimensions = 3601LL, 1801LL ;
                surface_net_solar_radiation:crs = "EPSG:4326" ;
                surface_net_solar_radiation:crs_transform = 0.1, 0., -180.05, 0., -0.1, 90.05 ;
                surface_net_solar_radiation:scale_factor = 0.1 ;
                surface_net_solar_radiation:bounds = 108.5, 28.75, 116.25, 33.5603826424351 ;
[...]

see details at JuliaGeo/NCDatasets.jl#265

[tylere]

Xee's .to_netcdf() method currently does not use the dimension ordering that is expected by many Python plotting libraries, and that appears to include the R terra package.

I started a discussion on whether to update the dimension ordering: #196
Please add your thoughts on whether the ordering should be a change (which would likely be a breaking change for some Xee users).