Extent outside of TMS bounds

[AndrewAnnex]

I am converting a dataset of geotiffs to web-optimized COGs with rio-cogeo using a TMS defined within the planetcantile project. The dataset consists of a large number of non-COG 4x4 degree tiles spanning -180 to 180 degrees and +/- 88 latitude, and have naming convention related to the lower left point, so tile E-180_N00 spans -180 to -176 longitude and 0 to 4 degrees latitude. Each tile has a 5x5 meter per pixel resolution.

I am specifically using the custom TMS from planetcantile for the IAU projection IAU_2015:49910 (https://www.opengis.net/def/crs/IAU/2015/49910, https://spatialreference.org/ref/iau_2015/49910/) that pyproj supports as well as recent rasterio releases.

The left-side boundary easting is -10669445.554195119, this will be important in a few more sentences...

Each raw TIF file is around 2 Gb in size, so I was surprised when a few COGs processed with rio-cogeo produced COGs that were only a few hundred Kb in size.

Specifically it was all of the tiles with a origin point on the antimeridian (-180 degrees longitude) except for the "center" tile on the equator (E-180_N00).

After inspecting the metadata on these tiles, it appears that it is a floating precision issue with the origin point X point. The origin easting for the E-180_N00 tile is -10669445.554197242483497 while the next tile above it at E-180_N04 has an origin of -10669445.554209999740124. The N04 tif also ends up wrapping all the way over to +180 degrees east, so although both origins are technically outside of the projection's extent, it seems that the E-180_N04 origin is large enough to be a problem.

While I will probably go through the exercise of using gdal_edit.py to modify the source Tif files to fix this (and I have already observed this fixing the issue of the resulting COG from rio cogeo, I wonder if there is a simple fix in rio-cogeo to make to avoid this extra step?

I would expect this to have something to do with the WarpedVRT usage and the lack of boundless reading, and I saw a related PR in rio-tiler that maybe could inspire a fix, but I haven't dug into this deeply enough to know for sure.

I made faux files for the inputs and outputs from rio-cogeo with the _tms.tif postfix, and the source files are prefixed by c_. Below are the info's for the COG outputs that shows that the N04 tif has a width that doesn't make sense and world spanning extent.

faux_E-180_N00_tms.tif info:

Driver: GTiff
File: /scratch/aannex/antimeridian_issue/faux_E-180_N00_tms.tif
COG: True
Compression: ZSTD
ColorSpace: None

Profile
    Width:            46848
    Height:           46848
    Bands:            1
    Tiled:            True
    Dtype:            uint8
    NoData:           0.0
    Alpha Band:       False
    Internal Mask:    False
    Interleave:       BAND
    ColorMap:         False
    ColorInterp:      ('gray',)
    Scales:           (1.0,)
    Offsets:          (0.0,)

Geo
    Crs:              PROJCS["Mars (2015) - Sphere / Equidistant Cylindrical",GEOGCS["Mars (2015) - Sphere",DATUM["Mars (2015) - Sphere",SPHEROID["Mars (2015) - Sphere",3396190,0]],PRIMEM["Reference Meridian",0],UNIT["degree",0.0174532925199433,AUTHORITY["EPSG","9122"]]],PROJECTION["Equirectangular"],PARAMETER["standard_parallel_1",0],PARAMETER["central_meridian",0],PARAMETER["false_easting",0],PARAMETER["false_northing",0],UNIT["metre",1,AUTHORITY["EPSG","9001"]],AXIS["Easting",EAST],AXIS["Northing",NORTH]]
    Origin:           (-10669445.554195119, 238343.3271994274)
    Resolution:       (5.087588097665367, -5.087588097665367)
    BoundingBox:      (-10669445.554195119, 3.2014213502407074e-10, -10431102.226995692, 238343.3271994274)
    MinZoom:          None
    MaxZoom:          None

Image Metadata
    AREA_OR_POINT: Area

Image Structure
    LAYOUT: COG
    COMPRESSION: ZSTD
    INTERLEAVE: BAND

Band 1
    ColorInterp: gray

IFD
    Id      Size           BlockSize     Decimation           
    0       46848x46848    256x256       0
    1       23424x23424    256x256       2
    2       11712x11712    256x256       4
    3       5856x5856      256x256       8
    4       2928x2928      256x256       16
    5       1464x1464      256x256       32
    6       732x732        256x256       64
    7       366x366        256x256       128
    8       183x183        256x256       256

faux_E-180_N04_tms.tif info:

Driver: GTiff
File: /scratch/aannex/antimeridian_issue/faux_E-180_N04_tms.tif
COG: True
Compression: ZSTD
ColorSpace: None

Profile
    Width:            65536
    Height:           1024
    Bands:            1
    Tiled:            True
    Dtype:            uint8
    NoData:           0.0
    Alpha Band:       False
    Internal Mask:    False
    Interleave:       BAND
    ColorMap:         False
    ColorInterp:      ('gray',)
    Scales:           (1.0,)
    Offsets:          (0.0,)

Geo
    Crs:              PROJCS["Mars (2015) - Sphere / Equidistant Cylindrical",GEOGCS["Mars (2015) - Sphere",DATUM["Mars (2015) - Sphere",SPHEROID["Mars (2015) - Sphere",3396190,0]],PRIMEM["Reference Meridian",0],UNIT["degree",0.0174532925199433,AUTHORITY["EPSG","9122"]]],PROJECTION["Equirectangular"],PARAMETER["standard_parallel_1",0],PARAMETER["central_meridian",0],PARAMETER["false_easting",0],PARAMETER["false_northing",0],UNIT["metre",1,AUTHORITY["EPSG","9001"]],AXIS["Easting",EAST],AXIS["Northing",NORTH]]
    Origin:           (-10669445.554195119, 500130.2603528965)
    Resolution:       (325.60563825058347, -325.60563825058347)
    BoundingBox:      (-10669445.554195119, 166710.08678429906, 10669445.554195119, 500130.2603528965)
    MinZoom:          None
    MaxZoom:          None

Image Metadata
    AREA_OR_POINT: Area

Image Structure
    LAYOUT: COG
    COMPRESSION: ZSTD
    INTERLEAVE: BAND

Band 1
    ColorInterp: gray

IFD
    Id      Size           BlockSize     Decimation           
    0       65536x1024     256x256       0
    1       32768x512      256x256       2
    2       16384x256      256x256       4

faux_src.zip

faux_tms.zip