CPLE_AppDefinedError: Unable to compute a RESAMPLING.NEAREST based transformation

[arthur-e]

Expected behavior and actual behavior.

I expected reproject() to run without error. Instead, I received the error:

---------------------------------------------------------------------------
CPLE_AppDefinedError                      Traceback (most recent call last)
Cell In [27], line 16
     12 destination = rio.open(
     13     '', 'w+', driver = 'MEM', crs = ref_ds.crs, transform = ref_ds.transform,
     14     width = ref_ds.width, height = ref_ds.height, count = 1, dtype = np.float32)
     15 source.write(red, 1)
---> 16 reproject(
     17     source = rio.band(source, 1),
     18     destination = rio.band(destination, 1),
     19     method = Resampling.nearest)
     20 red_proj = destination.read(1)

File /usr/local/python-env/gis/lib/python3.10/site-packages/rasterio/env.py:394, in ensure_env.<locals>.wrapper(*args, **kwds)
    392 else:
    393     with Env.from_defaults():
--> 394         return f(*args, **kwds)

File /usr/local/python-env/gis/lib/python3.10/site-packages/rasterio/warp.py:344, in reproject(source, destination, src_transform, gcps, rpcs, src_crs, src_nodata, dst_transform, dst_crs, dst_nodata, dst_resolution, src_alpha, dst_alpha, resampling, num_threads, init_dest_nodata, warp_mem_limit, **kwargs)
    340         destination = np.empty((int(dst_count), int(dst_height), int(dst_width)),
    341                                dtype=source.dtype)
    343 # Call the function in our extension module.
--> 344 _reproject(
    345     source, destination, src_transform=src_transform, gcps=gcps, rpcs=rpcs,
    346     src_crs=src_crs, src_nodata=src_nodata, dst_transform=dst_transform,
    347     dst_crs=dst_crs, dst_nodata=dst_nodata, dst_alpha=dst_alpha,
    348     src_alpha=src_alpha, resampling=resampling,
    349     init_dest_nodata=init_dest_nodata, num_threads=num_threads,
    350     warp_mem_limit=warp_mem_limit, **kwargs)
    352 return destination, dst_transform

File rasterio/_warp.pyx:516, in rasterio._warp._reproject()

File rasterio/_warp.pyx:488, in rasterio._warp._reproject()

File rasterio/_err.pyx:221, in rasterio._err.exc_wrap_pointer()

CPLE_AppDefinedError: Unable to compute a RESAMPLING.NEAREST based transformation between pixel/line and georeferenced coordinates for .

Steps to reproduce the problem.

The intent was to project one Landsat raster dataset (from Landsat 9) to match another (from Landsat 8). The data were downloaded from USGS Earth Explorer.

filename1 = f'{LANDSAT_DIR}/LC08_L2SP_015033_20220601_20220609_02_T1_SR_B4.TIF'
filename2 = f'{LANDSAT_DIR}/LC09_L2SP_015033_20220625_20230409_02_T1_SR_B4.TIF'

# Both files have the same CRS but different affine transformations
ref_ds = rio.open(filename1) # File I want to match
target_ds = rio.open(filename2) # File I want to change to match the other
original_transform = target_ds.transform
red = target_ds.read(1)

source = rio.open(
    '', 'w+', driver = 'MEM', crs = ref_ds.crs, transform = original_transform,
    width = red.shape[1], height = red.shape[0], count = 1, dtype = np.float32)
destination = rio.open(
    '', 'w+', driver = 'MEM', crs = ref_ds.crs, transform = ref_ds.transform,
    width = ref_ds.width, height = ref_ds.height, count = 1, dtype = np.float32)

# To facilitate a minimal working example... I'm actually doing a lot to the array
#    before writing it into the "source" dataset
source.write(red, 1)
reproject(
    source = rio.band(source, 1),
    destination = rio.band(destination, 1),
    method = Resampling.nearest)
red_proj = destination.read(1)

Environment Information

• Python 3.10.6
• rasterio 1.3.4
• GDAL 3.4.1
• Linux-5.19.0-41-generic-x86_64-with-glibc2.35

Installation Method

Within a virtual environment:

pip install rasterio

[groutr]

It will be helpful if you have a link to these two files or files that are similar. The error seems to be raised from inside GDAL. Does the error occur with other resampling methods? What are the CRS and transforms for these two files?

[arthur-e]

The two Landsat files can be downloaded here and here. The error does occur with bilinear resampling as well.

>>> ref_ds.transform
Affine(30.0, 0.0, 220485.0,
       0.0, -30.0, 4426815.0)

>>> target_ds.transform
Affine(30.0, 0.0, 215985.0,
       0.0, -30.0, 4426515.0)

>>> ref_ds.crs
CRS.from_epsg(32618)

>>> target_ds.crs
CRS.from_epsg(32618)

[sgillies]

I've moved this to a discussion because it looks like a usage question to me. @arthur-e what is your intent with source = rio.open('', 'w+', driver = 'MEM', crs = ref_ds.crs, transform = original_transform, width = red.shape[1], height = red.shape[0], count = 1, dtype = np.float32)? That dataset doesn't have any data and can't be used as a source for reprojection.

[arthur-e]

I write data to source on this line:

# To facilitate a minimal working example... I'm actually doing a lot to the array
#    before writing it into the "source" dataset
source.write(red, 1)

Which is prior to calling reproject(), so I would expect that there are source data to work with. But perhaps I am missing a step to "flush" the cache or something like that?

[sgillies]

@arthur-e I think that's it! GDAL has a raster block cache. source.write() puts data in the cache. Data only goes to disk when the cache is full or when the dataset is closed. I recommend that you close source after writing to it and then re-open it before you reproject.