Output raster of rasterio.mask doesn't match CRS of input [closed]

Question

I'm trying to mask a raster with population data with my area of interest. I reprojected the original population raster according to the AOI's crs, but after masking and writing the new file, the crs is set back to the original's crs, even though both inputs have the right crs (which is 32644). Can anyone point me to my mistake? I know I could reproject again, but there has to be a simpler way, right?

raster_crs = CRS(AOI.crs).to_epsg()

open original raster

with rio.open(pop_path, mode='r+') as pop:
    transform, width, height = calculate_default_transform(
        pop.crs, raster_crs, pop.width, pop.height, *pop.bounds)
    kwargs = pop.meta.copy()
    kwargs.update({
        'crs': raster_crs,
        'transform': transform,
        'width': width,
        'height': height
    })

CRS(pop.crs).to_epsg()

4326

reproject original raster to AOI's crs

    with rio.open(pop_proj_tif, 'w', **kwargs) as pop_proj:
        for i in range(1, pop.count + 1):
            reproject(
                source=rio.band(pop, i),
                destination=rio.band(pop_proj, i),
                pop_transform=pop.transform,
                pop_crs=pop.crs,
                pop_proj_transform=transform,
                pop_proj_crs=raster_crs,
                resampling=Resampling.nearest)

CRS(AOI.crs).to_epsg()

32644

CRS(pop_proj.crs).to_epsg()

32644

mask reprojected raster with AOI

with rio.open(pop_proj_tif) as pop_proj:
    pop_out_image, out_transform = rio.mask.mask(pop_proj, AOI.geometry, crop=True)
    pop_out_meta = pop.meta

pop_out_meta.update({"driver": "GTiff",
                 "height": pop_out_image.shape[1],
                 "width": pop_out_image.shape[2],
                     "transform": out_transform})
    

writing masked raster gives wrong CRS

    with rio.open(pop_proj_clip_tif, "w", **pop_out_meta) as pop_dest:
        pop_dest.write(pop_out_image)

CRS(pop_dest.crs).to_epsg()

4326

Answer 1

You are copying pop.meta instead of pop_proj.meta.

Replace:

# mask reprojected raster with AOI
with rio.open(pop_proj_tif) as pop_proj:
    pop_out_image, out_transform = rio.mask.mask(pop_proj, AOI.geometry, crop=True)
    pop_out_meta = pop.meta #<== wrong

With:

# mask reprojected raster with AOI
with rio.open(pop_proj_tif) as pop_proj:
    pop_out_image, out_transform = rio.mask.mask(pop_proj, AOI.geometry, crop=True)
    pop_out_meta = pop_proj.meta #<== this works

Also the reproject function signature doesn't have the following arguments - pop_transform, pop_crs, pop_proj_transform or pop_proj_crs and it's only a lucky accident that you didn't get a TypeError: reproject() got an unexpected keyword argument 'pop_transform' as the reproject function accepts a **kwargs argument for any additional arguments that should be passed to both the image to image transformer GDALCreateGenImgProjTransformer2() (for example, MAX_GCP_ORDER=2) and the GDALWarpOptions (for example, INIT_DEST=NO_DATA).

Use the following instead:

        reproject(
            source=rio.band(pop, i),
            destination=rio.band(pop_proj, i),
            src_transform=pop.transform,
            src_crs=pop.crs,
            dst_transform=transform,
            dst_crs=raster_crs,
            resampling=Resampling.nearest)

Or just drop them completely as they will be read from the source and dest bands anyway:

        reproject(
            source=rio.band(pop, i),
            destination=rio.band(pop_proj, i),
            resampling=Resampling.nearest)