Gdalwarp producing multiple copies of input image at incorrect positions

[bch4132]

What is the bug?

Gdalwarp is producing artifacts in output when using -te arguments larger than bounds of the input

Steps to reproduce the issue

repro.zip
Inside the zip attached there is tiff file used as input. When running this through gdal with the following arguments:
-t_srs "+proj=longlat +datum=WGS84 +ellps=WGS84 +no_defs +type=crs" -of GTiff -co "TILED=YES" -co "BLOCKXSIZE=1024" -co "BLOCKYSIZE=1024" -co "BIGTIFF=YES" -wm 500 -wo "NUM_THREADS=ALL_CPUS" -multi -co "COMPRESS=ZSTD" -co "NUM_THREADS=ALL_CPUS" -dstnodata -32767.0 -ot Float32 -te -81.500000000000 28.500000000000 -81.000000000000 29.000000000000 -r near "n4.tif" "Repro_hgt.tif"
There are copies of the input produced on the left and right side of where the actual input should be on the resulting image.

Input image:

Resulting output:

Versions and provenance

I have tested with multiple versions, including 3.6.2, 3.9.2 (from QGIS install), as well as 3.10.0dev-d04e73dfbd from a source build all with the same result.

Additional context

No response

[jratike80]

I could re-produce on Windows with GDAL 3.9.1. Without -te the output is good and corner coordinates are:

Corner Coordinates:

Upper Left  ( -81.1762512,  28.6631436) ( 81d10'34.50"W, 28d39'47.32"N)
Lower Left  ( -81.1762512,  28.6354970) ( 81d10'34.50"W, 28d38' 7.79"N)
Upper Right ( -81.1317603,  28.6631436) ( 81d 7'54.34"W, 28d39'47.32"N)
Lower Right ( -81.1317603,  28.6354970) ( 81d 7'54.34"W, 28d38' 7.79"N)
Center      ( -81.1540058,  28.6493203) ( 81d 9'14.42"W, 28d38'57.55"N)

What is this coordinate system of n4.tif?

Coordinate System is:
GEOGCRS["World Geodetic System 1984",
    DATUM["World Geodetic System 1984",
        ELLIPSOID["WGS 84",6378137,298.257223563,
            LENGTHUNIT["metre",1]],
        ID["EPSG",6326]],
    PRIMEM["Greenwich",0,
        ANGLEUNIT["arc-second",4.84813681109536E-06]],
    CS[ellipsoidal,2],
        AXIS["latitude",north,
            ORDER[1],
            ANGLEUNIT["arc-second",4.84813681109536E-06]],
        AXIS["longitude",east,
            ORDER[2],
            ANGLEUNIT["arc-second",4.84813681109536E-06]]]

[jratike80]

I think that the issue must have something to do with those arc-seconds. If I first warp the image into EPSG:4326 without any parameters, and then this interim image into the target extent, then the result looks good.

I have removed all parameters which I thought are not essential.

gdalwarp -t_srs epsg:4326 -te -81.500000000000 28.500000000000 -81.000000000000 29.000000000000 "repro_interim_4326.tif" "Repro_4.tif"

[rouault]

I think that the issue must have something to do with those arc-seconds.

Exactly. Fix in #10988