Fixing the GOES transformation to regular lat/lon COG

[j08lue]

So the conversion of GOES from NetCDF to COG did not produce a correctly geo-referenced GeoTIFF with

https://github.com/NASA-IMPACT/veda-cyclone/blob/56037d536aa801fa3f6a767b28e9c8c53e80b4de/transformation_script/convert_netcdf_cog_GOES.ipynb

The issue was that rioxarray/rasterio/GDAL does not work well with 2D coordinate variables created from transforming the GOES coordinates to lat/lon.

Instead, you need to reproject the data to a regular lat/lon grid.

This works and created correctly geo-referenced COGs:

from pyproj import CRS

def goes_to_wgs(ds, variable_name):
    
    # convert coordinates to meters https://gis.stackexchange.com/a/350006
    sat_height = ds["goes_imager_projection"].attrs["perspective_point_height"]
    ds = ds.assign_coords({
        "x": ds["x"].values * sat_height,
        "y": ds["y"].values * sat_height,
    })
    
    crs = CRS.from_cf(ds["goes_imager_projection"].attrs)
    ds.rio.write_crs(crs.to_string(), inplace=True)
    
    da = ds[variable_name]

    return da.rio.reproject("epsg:4326")

Create a COG from a NetCDF file on disk:

curl -O https://noaa-goes16.s3.amazonaws.com/ABI-L1b-RadF/2017/129/11/OR_ABI-L1b-RadF-M3C03_G16_s20171291115392_e20171291126159_c20171291126196.nc

import xarray as xr
import rioxarray

path = "OR_ABI-L1b-RadF-M3C03_G16_s20171291115392_e20171291126159_c20171291126196.nc"

with xr.open_dataset(path, engine='h5netcdf') as ds:
    da = goes_to_wgs(ds, variable_name="Rad")
    COG_PROFILE = {"driver": "COG", "compress": "DEFLATE", "predictor": 2}
    da.rio.to_raster("goes16.tif", **COG_PROFILE)

or straight from AWS S3:

import s3fs 
import xarray as xr
import rioxarray

fs = s3fs.S3FileSystem(anon=True)

path = "s3://noaa-goes16/ABI-L1b-RadF/2017/129/11/OR_ABI-L1b-RadF-M3C03_G16_s20171291115392_e20171291126159_c20171291126196.nc"

with fs.open(path) as fileObj:
    with xr.open_dataset(fileObj, engine='h5netcdf') as ds:
        da = goes_to_wgs(ds, variable_name="Rad")
        COG_PROFILE = {"driver": "COG", "compress": "DEFLATE", "predictor": 2}
        da.rio.to_raster("goes16.tif", **COG_PROFILE)

Notebook here: https://gist.github.com/j08lue/f949404cf95c9a521b6b72f33a4cbf3b