Artefacts using WebMercatorQuad TMS

[vprint]

I'm trying to create COG of french LIDAR HD DEM.

While cog_trranslate produces good looking COG (see screen 1), the tms=morecantile.tms.get("WebMercatorQuad") parameters create strange artefacts (see screen 2). Do you have any idea on how to fix this ?

Here is the script i use:

def export_to_cog(
    src_path: Path | str,
    dst_path: Path | str,
    profile: COGProfile = COGProfile.DEFLATE,
    profile_options: Optional[dict] = None,
    **options,
) -> None:
    """
    Export a raster to COG.

    Args:
        src_path: The source raster path.
        dst_path: The target COG path.
        profile: The compression profile.
        profile_options: The compression options.
        **options: Additional rio-cogeo options (see cog_translate).
    """
    output_profile = cog_profiles.get(profile.value)
    output_profile.update(dict(BIGTIFF="IF_SAFER"))

    if profile_options:
        output_profile.update(profile_options)

    config = dict(
        GDAL_NUM_THREADS="ALL_CPUS",
        GDAL_TIFF_INTERNAL_MASK=True,
        GDAL_TIFF_OVR_BLOCKSIZE="128",
    )

    cog_translate(
        src_path,
        dst_path,
        output_profile,
        config=config,
        in_memory=False,
        **options,
    )

I call this method like this:

export_to_cog(
    f"{VRT_FOLDER}/raw_datas.vrt",
    f"{WORKING_SANCY}/cog_raw_tms.tif",
    profile=COGProfile.LERC_ZSTD,
    # profile_options={"MAX_Z_ERROR": 0.01},
    tms=morecantile.tms.get("WebMercatorQuad"),
)

COG without tms=morecantile.tms.get("WebMercatorQuad")

Artefacts with tms=morecantile.tms.get("WebMercatorQuad")

[vincentsarago]

Hi @vprint

Can you try using other resampling method than nearest?

export_to_cog(
    f"{VRT_FOLDER}/raw_datas.vrt",
    f"{WORKING_SANCY}/cog_raw_tms.tif",
    profile=COGProfile.LERC_ZSTD,
    # profile_options={"MAX_Z_ERROR": 0.01},
    tms=morecantile.tms.get("WebMercatorQuad"),
    resampling="bilinear",
    overview_resampling="bilinear",
)

[vprint]

Hi, it's much better with bilinear. Thanks !